education-agent-skills

Adaptive Hint Sequence Designer

What This Skill Does

Designs a cascading hint sequence for a specific problem type — a series of progressively more revealing hints that help students move past sticking points without simply giving them the answer. This is one of the most technically demanding aspects of intelligent tutoring system (ITS) design. The critical insight from VanLehn's (2011) meta-analysis is that the effectiveness of tutoring (human or AI) depends heavily on the quality of the scaffolding — and hint sequences are the primary scaffolding mechanism. Get the sequence wrong and you either frustrate students (hints too vague, too few) or steal their learning (hints too specific, too early). The output includes the complete hint cascade (typically 3–5 levels from general strategic guidance to specific procedural nudge), a design rationale explaining the cognitive function of each level, trigger conditions (when each hint fires), and a bottom-out strategy (what happens when hints are exhausted). AI is specifically valuable here because designing effective hint sequences requires simultaneously anticipating student errors, calibrating hint specificity, and ensuring that each hint level provides just enough information to unstick the student without bypassing the cognitive work that produces learning.

Evidence Foundation

VanLehn (2011) conducted the most comprehensive meta-analysis of tutoring effectiveness, comparing human tutoring, intelligent tutoring systems, and other approaches. He found ITS effect sizes averaging 0.76 — remarkably close to human tutoring (0.79) and substantially higher than "no tutoring" conditions. Critically, ITS effectiveness depended on the quality of the step-level interaction: systems that provided feedback and hints at each problem-solving step (inner loop) were much more effective than systems that only evaluated the final answer (outer loop). Aleven & Koedinger (2002) studied hint-seeking behaviour in the Carnegie Learning Cognitive Tutor and found that students often used hints suboptimally — either requesting hints too quickly (before attempting the problem) or too slowly (struggling unproductively). They found that training students in a metacognitive hint strategy ("try first, then ask for a hint, then explain the hint to yourself") significantly improved learning outcomes. Razzaq & Heffernan (2010) compared proactive hints (given automatically) with reactive hints (given on request) and found that the optimal approach depended on student proficiency: lower-performing students benefited more from proactive hints, while higher-performing students benefited from being allowed to struggle before requesting help. Shute (2008) reviewed formative feedback research and identified that effective feedback is specific, timely, and actionable — principles that apply directly to hint design. Wood, Bruner & Ross (1976) established the concept of scaffolding: providing temporary support that enables the learner to accomplish what they cannot do alone, then gradually withdrawing the support as competence develops.

Input Schema

The teacher must provide:

• Problem type: What students are solving. e.g. "Solving linear equations with one unknown — e.g. 3x + 7 = 22" / "Writing a topic sentence for a persuasive paragraph" / "Balancing a chemical equation" / "Debugging a Python function that should return a sorted list but returns None"
• Common sticking points: Where students get stuck. e.g. "Students forget to do the same operation to both sides of the equation" / "Students write topic sentences that are too vague or that state a fact rather than a claim" / "Students balance atoms randomly rather than systematically" / "Students don't understand that Python's sort() returns None — they expect it to return the sorted list"

Optional (injected by context engine if available):

• Student level: Year group and proficiency
• Subject area: The curriculum subject
• Delivery context: AI system, teacher, or materials
• Number of hint levels: How many levels to design
• Final hint policy: What happens at the bottom of the sequence

Prompt

You are an expert in hint sequence design for intelligent tutoring systems, with deep knowledge of VanLehn's (2011) meta-analysis of ITS effectiveness, Aleven & Koedinger's (2002) research on hint-seeking behaviour, Razzaq & Heffernan's (2010) work on proactive vs. reactive hints, Shute's (2008) formative feedback principles, and Wood, Bruner & Ross's (1976) scaffolding framework. You understand that hint design is the hardest part of ITS design: too little help produces frustration and abandonment; too much help produces answer-copying without learning. The optimal hint sequence progressively reduces the cognitive work required while preserving as much productive struggle as possible.

CRITICAL PRINCIPLES:
- **Progressive specificity.** The first hint should be the MOST GENERAL — pointing the student toward the right strategy or concept without revealing the specific step. Each subsequent hint should be more specific, narrowing the cognitive search space while still requiring the student to do some thinking. The final hint (the "bottom-out hint") may reveal the specific step but should still require the student to execute it and understand why.
- **Each hint should do ONE thing.** A hint that says "Remember to isolate the variable by subtracting 7 from both sides and then dividing by 3" is not a hint — it's the answer in disguise. Each hint level should address ONE aspect of the sticking point.
- **Hints should address the CAUSE of the error, not just the symptom.** If a student gets 3x + 7 = 22 wrong, the hint should not just say "Try again" (too vague) or "x = 5" (the answer). It should address the likely CAUSE: "What operation will move the 7 to the other side?" — targeting the specific conceptual gap.
- **Trigger conditions matter as much as hint content.** When a hint fires is as important as what it says. Too early: the student doesn't get the chance to struggle productively. Too late: the student has given up. Design triggers based on observable behaviour: time elapsed, number of incorrect attempts, specific error patterns, or student request.
- **The bottom-out strategy must still produce learning.** When all hints are exhausted, the system cannot just shrug and move on. The bottom-out should provide the answer WITH a self-explanation prompt: "The answer is X because Y. In your own words, explain why Y leads to X." This converts the bottom-out from answer-giving to learning.

Your task is to design a hint sequence for:

**Problem type:** {{problem_type}}
**Common sticking points:** {{common_sticking_points}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Student level:** {{student_level}} — if not provided, design for a general secondary school context.
**Subject area:** {{subject_area}} — if not provided, infer from the problem type.
**Delivery context:** {{delivery_context}} — if not provided, design for an AI tutoring system that can also be used by teachers.
**Number of hint levels:** {{number_of_hint_levels}} — if not provided, design 4 levels (strategic → conceptual → procedural → bottom-out).
**Final hint policy:** {{final_hint_policy}} — if not provided, design a bottom-out hint with self-explanation prompt.

Return your output in this exact format:

## Adaptive Hint Sequence: [Problem Type]

**Problem type:** [What students are solving]
**Common sticking points:** [Where they get stuck]
**Hint levels:** [How many levels, from general to specific]

### Hint Architecture

[Overview of the hint strategy — what each level does and why the sequence is ordered this way]

### Hint Sequence

For each level:
**Level [N]: [Hint type — Strategic / Conceptual / Procedural / Bottom-out]**
- **Hint text:** [Exactly what the student sees or hears]
- **Cognitive function:** [What this hint does — what cognitive work it performs for the student and what it preserves for them to do]
- **Trigger condition:** [When this hint fires — time, errors, request, behaviour]
- **Expected student response:** [What a student who benefits from this hint will do next]
- **If this hint doesn't help:** [What happens — escalate to next level]

### Bottom-Out Strategy

[What happens when all hints are exhausted — the specific design for revealing the answer while still producing learning]

### Trigger Conditions Summary

| Trigger Type | Condition | Hint Level Activated |
|---|---|---|
| [Type] | [Specific condition] | [Which level] |

### Adaptation Notes

[How the sequence adapts to different student profiles — stronger students may skip early levels; weaker students may need the full cascade]

**Self-check before returning output:** Verify that (a) hints are progressively more specific, (b) each hint does ONE thing, (c) hints address the cause of the error not just the symptom, (d) trigger conditions are specified, (e) the bottom-out still produces learning, and (f) the sequence preserves productive struggle.

Example Output

Scenario: Problem type: "Solving linear equations with one unknown — specifically equations of the form ax + b = c, where students must isolate x. Example: 3x + 7 = 22" / Common sticking points: "Students forget to perform the same operation on both sides (they subtract 7 from the left but not the right). Students don't understand the inverse operation concept (they add when they should subtract, or vice versa). Students attempt operations in the wrong order (dividing by 3 before subtracting 7)." / Delivery context: "AI tutoring system — hints appear on screen when triggered" / Student level: "Year 8, first term of algebra"

---

Adaptive Hint Sequence: Solving Linear Equations (ax + b = c)

Problem type: Solving linear equations with one unknown — isolating x in equations like 3x + 7 = 22 Common sticking points: Same-operation-both-sides errors, inverse operation confusion, wrong operation order Hint levels: 4 (strategic → conceptual → procedural → bottom-out)

Hint Architecture

The sequence moves from STRATEGIC (pointing toward an approach) through CONCEPTUAL (activating the relevant principle) to PROCEDURAL (indicating the specific step) to BOTTOM-OUT (revealing the step with self-explanation). At each level, the student retains some cognitive work:

• Level 1 (Strategic): "Here's the GOAL" — the student must figure out the approach
• Level 2 (Conceptual): "Here's the PRINCIPLE" — the student must figure out the specific operation
• Level 3 (Procedural): "Here's the STEP" — the student must execute and verify
• Level 4 (Bottom-out): "Here's the ANSWER for this step" — the student must explain why

Hint Sequence

Level 1: Strategic — "What's your goal?"

• Hint text: "Your goal is to get x by itself on one side of the equation. Right now, x has some extra numbers attached to it. What could you do to start removing them?"
• Cognitive function: Reminds the student of the overall strategy (isolate x) without indicating which specific operation to perform or which number to deal with first. The student must still identify which numbers to remove and which operations to use.
• Trigger condition: Student has made no progress for 60 seconds, OR has entered an incorrect value for x without showing working, OR has requested a hint.
• Expected student response: Student identifies that they need to "get rid of" the +7 or the 3, and attempts an operation. Even if the operation is wrong, the strategic framing redirects their attention to the goal.
• If this hint doesn't help: After one more incorrect attempt or 45 seconds of no progress, escalate to Level 2.

Level 2: Conceptual — "Which number first, and what's its opposite?"

• Hint text: "Look at the equation: 3x + 7 = 22. The x is being multiplied by 3 AND has 7 added to it. Deal with the ADDITION first. What is the opposite of adding 7?"
• Cognitive function: This hint does TWO things: it tells the student the ORDER (deal with addition before multiplication — the reverse of BIDMAS/PEMDAS) and it activates the INVERSE OPERATION concept (the opposite of adding). The student must still figure out that "the opposite of adding 7 is subtracting 7" and apply it to both sides.
• Trigger condition: Student has attempted an incorrect operation after Level 1 (e.g., dividing by 3 first), OR has correctly identified "subtract" but not applied it to both sides.
• Expected student response: Student writes 3x + 7 - 7 = 22 - 7, simplifying to 3x = 15. If they subtract from only one side, Level 3 addresses this.
• If this hint doesn't help: After one more error or 30 seconds, escalate to Level 3.

Level 3: Procedural — "Do the same thing to both sides"

• Hint text: "Subtract 7 from BOTH sides of the equation. The left side becomes 3x + 7 - 7 = 3x. What does the right side become? 22 - 7 = ?"
• Cognitive function: This hint specifies the exact operation (subtract 7) and emphasises the both-sides principle. The student must still calculate 22 - 7 = 15 and then figure out the next step (dividing by 3). The hint solves the FIRST step but leaves the SECOND step for the student.
• Trigger condition: Student has subtracted from only one side, or has been unable to identify the correct operation after Level 2.
• Expected student response: Student writes 3x = 15, then attempts to isolate x by dividing both sides by 3, arriving at x = 5.
• If this hint doesn't help: If student still cannot complete the problem after this hint (cannot calculate 22 - 7, or does not know to divide by 3 next), escalate to Bottom-out.

Level 4: Bottom-out — "Here's the step. Explain why."

• Hint text: "Here's the full solution: 3x + 7 = 22 → Subtract 7 from both sides → 3x = 15 → Divide both sides by 3 → x = 5. Now: in your own words, explain WHY we subtracted 7 before dividing by 3. What would have happened if we divided by 3 first?"
• Cognitive function: The answer is fully revealed, but the self-explanation prompt preserves learning. By asking "why this order?" the system forces the student to engage with the mathematical logic rather than just copying the answer. The counter-factual question ("what if we divided first?") develops deeper understanding of operation order.
• Trigger condition: All previous hints exhausted and student still cannot complete the problem.
• Expected student response: Student explains (in text or selection): "We subtract 7 first because it's easier / because the 7 is being added, not multiplied by x / because if we divided by 3 first, we'd have to divide the 7 as well, which gives fractions." Any of these indicates understanding.

Bottom-Out Strategy

The bottom-out hint reveals the complete solution but requires a self-explanation before the student can proceed to the next problem. The system presents two prompts:

1. Explain the solution: "Why did we subtract 7 before dividing by 3?" (free text or multiple choice)
2. Predict a similar problem: "If the equation were 5x + 12 = 37, what would you do first?" (tests whether the student can transfer the principle to a new instance)

If the student's self-explanation is accurate, proceed to the next problem (a similar equation, to test immediate application). If the self-explanation is inaccurate ("because 7 is smaller than 22"), provide a brief corrective explanation and then proceed. The bottom-out is NOT a failure state — it is a learning event. The self-explanation converts answer-giving into understanding-building.

Trigger Conditions Summary

Trigger Type	Condition	Hint Level Activated
Time-based	60 seconds with no progress	Level 1
Error-based	Incorrect final answer with no working shown	Level 1
Error-based	Incorrect operation (e.g., dividing before subtracting)	Level 2
Error-based	Correct operation applied to only one side	Level 3
Request-based	Student clicks "hint" button	Next level in sequence
Exhaustion	All previous hints delivered, still stuck	Level 4 (Bottom-out)

Adaptation Notes

For stronger students: If a student has successfully solved 3+ similar equations, skip Level 1 (they know the strategy). Start at Level 2 if they make an error — their sticking point is likely conceptual or procedural, not strategic.

For weaker students: If a student has needed the bottom-out hint on the previous 2 problems, consider switching from hints to a worked example — the student may lack the prerequisite knowledge that hints assume. Hints scaffold problem-solving; they don't teach foundational concepts.

For hint-abusers (Aleven & Koedinger, 2002): If a student rapidly clicks through hints without attempting the problem, insert a "try first" prompt before the first hint: "Have a go at the problem before asking for a hint. What's your best guess for the first step?" This addresses the metacognitive deficit: the student has learned to extract answers from hints rather than to think.

---

Known Limitations

1. Hint sequences assume the student has the prerequisite knowledge to benefit from hints. A hint that says "What's the opposite of adding 7?" assumes the student understands inverse operations. If they don't, no hint in this sequence will help — they need direct instruction on the prerequisite concept. Hint sequences scaffold PROBLEM-SOLVING, not KNOWLEDGE ACQUISITION. A student who lacks foundational knowledge needs a different intervention.

2. The optimal hint trigger is still an open research question. Razzaq & Heffernan (2010) found that proactive hints help weaker students and reactive hints help stronger students, but the field has not converged on a definitive trigger model. The trigger conditions above are reasonable defaults, not empirically optimised thresholds. In a real ITS, these would need to be calibrated through A/B testing with the actual student population.

3. Hint quality depends on accurate diagnosis of the error. The hint sequence above assumes specific error patterns (wrong operation order, one-sided operation, inverse confusion). If the student's actual error is different (e.g., an arithmetic mistake on 22 - 7), the hints will miss the mark. Effective hint systems need error-specific branching, not just linear cascades — which is significantly more complex to design.

AI-Facilitated Collaborative Learning Designer

What This Skill Does

Designs a collaborative learning task with specific AI facilitation points — the places where an AI system supports the group process without replacing it. This skill addresses the fundamental challenge of collaborative learning: it CAN be one of the most powerful learning approaches (Slavin's 1995 meta-analysis found effect sizes of 0.26-0.32 for well-structured cooperative learning) but it frequently degenerates into one student doing all the work while others watch, or into parallel individual work with a shared document. Dillenbourg (1999) established that genuine collaboration requires joint problem-solving with shared understanding, not just task division. Järvelä & Hadwin (2013) showed that effective collaboration requires socially shared regulation of learning (SSRL) — the group's ability to collectively plan, monitor, and adjust their approach. AI is specifically valuable here because it can do what teachers cannot: observe multiple groups simultaneously, detect collaboration breakdown in real time, and intervene precisely when needed. A teacher circulating among 8 groups catches problems minutes or hours late; an AI monitoring group interactions can prompt in real time. The output includes the complete collaboration design (task structure, roles, phases), AI facilitation moves (when and how the AI intervenes), regulation scaffolds (supporting the group's self-regulation), and equity mechanisms (ensuring all members participate).

Evidence Foundation

Dillenbourg (1999) established the foundational framework for computer-supported collaborative learning (CSCL), distinguishing between cooperation (dividing a task into subtasks that individuals complete separately) and collaboration (jointly constructing shared understanding of a problem). He argued that genuine collaboration requires: (a) a shared goal, (b) mutual engagement with each other's ideas, and (c) joint construction of knowledge that no individual could produce alone. Technology can support collaboration by structuring interaction, making thinking visible, and providing shared representations — but it can also undermine collaboration by making task division too easy. Roschelle & Teasley (1995) defined collaboration as "a coordinated, synchronous activity that is the result of a continued attempt to construct and maintain a shared conception of a problem." They showed that effective collaboration involves specific conversational patterns: proposals, elaborations, challenges, and repairs of shared understanding. When these patterns break down, collaboration degenerates into parallel work. Järvelä & Hadwin (2013) developed the concept of socially shared regulation of learning (SSRL) — the idea that effective collaborative groups don't just share the cognitive work, they also share the REGULATORY work: planning what to do, monitoring progress, evaluating whether their approach is working, and adjusting when it isn't. They found that SSRL is the strongest predictor of collaborative learning success, and that it can be scaffolded by technology — prompts that ask the group to plan, check progress, and reflect. Slavin (1995) conducted a comprehensive review of cooperative learning research, finding consistent positive effects (d = 0.26-0.32) when two conditions were met: (a) group goals (the group is assessed as a group, not individually) and (b) individual accountability (each member's contribution is visible and assessed). Without these conditions, cooperative learning often produces free-riding and social loafing. Kirschner et al. (2018) extended cognitive load theory to collaborative contexts, arguing that group work distributes cognitive load across members — but only when the task is too complex for any individual. For simple tasks, the transaction costs of collaboration (coordinating, communicating, managing different perspectives) outweigh the benefits. Collaboration should be reserved for tasks that genuinely require multiple minds.

Input Schema

The teacher must provide:

• Collaborative task: What students will work on together. e.g. "Year 10 science: design an experiment to test whether temperature affects the rate of a chemical reaction. Each group must produce a method, risk assessment, predicted results, and explanation of the underlying chemistry" / "Year 8 English: collaboratively write a newspaper front page covering the events of Romeo and Juliet as if they happened today" / "Year 12 economics: analyse a real-world case study of market failure and produce a policy brief recommending government intervention"
• Collaboration challenge: What goes wrong when students collaborate on this. e.g. "One student does all the writing while the others watch. The 'discussion' phase lasts 30 seconds before they start dividing up the task" / "Groups default to the loudest student's idea without considering alternatives. Quieter students disengage" / "Students split the work up ('you do the method, I'll do the risk assessment') and never build shared understanding — the final product is four separate pieces stuck together"

Optional (injected by context engine if available):

• Student level: Year group and proficiency
• Subject area: The curriculum subject
• Group size: How many students per group
• AI capabilities: What AI tools are available
• Time available: Duration of the collaborative task

Prompt

You are an expert in collaborative learning design and AI-facilitated CSCL (computer-supported collaborative learning), with deep knowledge of Dillenbourg's (1999) collaboration framework, Roschelle & Teasley's (1995) research on shared knowledge construction, Järvelä & Hadwin's (2013) socially shared regulation of learning (SSRL), Slavin's (1995) meta-analysis of cooperative learning, and Kirschner et al.'s (2018) collaborative cognitive load theory. You understand that collaboration is one of the most powerful — and most frequently misused — learning strategies. You know that the difference between productive collaboration and time-wasting group work depends entirely on the DESIGN, not on the students.

CRITICAL PRINCIPLES:
- **Collaboration requires a task that is genuinely too complex for one person.** Kirschner et al. (2018): if one student can do the task alone, collaboration adds transaction costs without cognitive benefit. The task must require multiple perspectives, skills, or knowledge bases that no individual possesses. If the task could be done individually, don't make it collaborative.
- **Structure the interaction, not just the task.** Dillenbourg (1999): giving students a task and saying "work together" is not collaborative learning design. The interaction must be structured: who speaks when, what they discuss at each phase, how they make decisions, and how they build on each other's ideas. AI can enforce this structure without the teacher being present.
- **Individual accountability prevents free-riding.** Slavin (1995): every group member must have a visible, assessable contribution. If a student can sit passively and still get the grade, they will. Design roles, individual deliverables, or accountability mechanisms that make every student's contribution visible.
- **AI should facilitate, not participate.** The AI's role is to support the GROUP PROCESS — prompting planning, flagging imbalanced participation, asking regulation questions ("How is your approach going? Do you need to adjust?"). The AI should NOT contribute ideas, solve problems for the group, or become a group member. If students are talking to the AI instead of to each other, the design has failed.
- **Socially shared regulation is the highest-value target.** Järvelä & Hadwin (2013): the groups that learn the most are the ones that collectively regulate their process — planning, monitoring, and adjusting. AI scaffolds for regulation (planning prompts, progress checks, reflection prompts) are more valuable than AI scaffolds for content.

Your task is to design an AI-facilitated collaborative learning experience:

**Collaborative task:** {{collaborative_task}}
**Collaboration challenge:** {{collaboration_challenge}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Student level:** {{student_level}} — if not provided, design for a general secondary school context.
**Subject area:** {{subject_area}} — if not provided, infer from the task.
**Group size:** {{group_size}} — if not provided, design for groups of 3-4.
**AI capabilities:** {{ai_capabilities}} — if not provided, design for a text-based AI that can monitor group work and send prompts to the group.
**Time available:** {{time_available}} — if not provided, design for a 50-minute lesson.

Return your output in this exact format:

## AI-Facilitated Collaboration Design: [Task Name]

**Task:** [What students will produce together]
**Collaboration challenge addressed:** [The specific problem this design solves]
**AI role:** [One-sentence summary of what the AI does and what it does NOT do]

### Task Complexity Check

[Verify that this task genuinely requires collaboration. If it doesn't, redesign it so it does.]

### Collaboration Structure

[The complete design: phases, roles, interaction rules, and deliverables. Include timing.]

**Phase 1: [Name]** (X minutes)
[What happens, what each role does, what the AI does]

**Phase 2: [Name]** (X minutes)
[What happens, what each role does, what the AI does]

### AI Facilitation Moves

[The specific moves the AI makes during collaboration — when, what, and why]

| Trigger | AI Move | Purpose | Example |
|---|---|---|---|
| [When] | [What the AI does] | [Why] | [Exact text] |

### Regulation Scaffolds

[How the AI supports socially shared regulation — planning, monitoring, and reflection prompts]

### Equity Mechanisms

[How the design ensures balanced participation — roles, accountability, and AI monitoring]

### What the AI Does NOT Do

[Explicit boundaries on AI involvement — what the AI should never do in this context]

**Self-check before returning output:** Verify that (a) the task genuinely requires collaboration, (b) interaction is structured not just assigned, (c) individual accountability is built in, (d) the AI facilitates the process rather than participating in it, and (e) regulation scaffolds are included.

Example Output

Scenario: Collaborative task: "Year 10 science: design an experiment to test whether temperature affects the rate of a chemical reaction (hydrochloric acid and sodium thiosulfate). Each group must produce a method, risk assessment, predicted results graph, and explanation of the underlying chemistry (collision theory)" / Collaboration challenge: "Groups always divide the work: one student writes the method, one does the risk assessment, one draws the graph, one does the theory. They never discuss, never build shared understanding, and the final report is four separate pieces stuck together. The student who wrote the method often doesn't understand the theory, and the student who wrote the theory couldn't replicate the method" / Group size: "4" / Student level: "Year 10, triple science" / AI capabilities: "Text-based AI chatbot that can message the group at set intervals or in response to triggers" / Time available: "60 minutes"

---

AI-Facilitated Collaboration Design: Experiment Design — Temperature and Reaction Rate

Task: Design a complete experiment to test whether temperature affects the rate of the hydrochloric acid / sodium thiosulfate reaction Collaboration challenge addressed: Task-splitting without shared understanding — students divide and conquer rather than collaborating, producing fragmented work AI role: The AI monitors group process and sends regulation prompts at key moments. It NEVER contributes scientific ideas, writes any part of the report, or answers chemistry questions.

Task Complexity Check

The original task CAN be divided into independent subtasks (method, risk assessment, graph, theory) — which is exactly why students split it up. To make collaboration necessary, the task must be REDESIGNED so that each component depends on the others:

Redesigned task: Instead of producing four separate sections, the group must produce ONE integrated document where every decision in the method is explicitly justified by the theory, every risk is linked to a specific step in the method, and the predicted results graph is derived from the theory (with the group explaining the shape of the curve). This interdependence means students CANNOT split the work — the student writing the method needs the theory student, and vice versa.

Additional constraint: Each group member must be able to explain ANY section of the report, not just "their" section. The teacher will randomly select one member at the end to present and answer questions about the entire report.

Collaboration Structure

Phase 1: Individual Thinking (5 minutes) Each student independently writes down: (a) what they already know about collision theory, (b) one thing they're unsure about, and (c) their prediction — will higher temperature increase or decrease the reaction rate, and why? This activates prior knowledge and gives every student something to contribute to the discussion. NO TALKING during this phase.

AI move: At the end of 5 minutes, the AI sends: "Time's up for individual thinking. Before you start discussing, each person shares ONE thing they wrote down. Go round the group — everyone speaks before anyone speaks twice."

Phase 2: Shared Understanding Building (15 minutes) The group discusses collision theory and how temperature affects reaction rate. They must reach AGREEMENT on: (a) why higher temperature increases the rate (particles move faster → more frequent AND more energetic collisions → more particles exceed activation energy), and (b) what the predicted results graph should look like (curved line showing rate increasing with temperature — not linear, because of the activation energy distribution). No writing yet — this is DISCUSSION only.

AI move: At minute 10, the AI sends a regulation check: "Quick check: can everyone in the group explain WHY higher temperature increases reaction rate? Not just 'particles move faster' — WHY does faster movement lead to a faster reaction? If anyone is unsure, this is the time to ask. You need this understanding for the next phase."

Phase 3: Integrated Design (25 minutes) The group writes the integrated report. To prevent task-splitting, use a ROTATING AUTHOR structure:

• Minutes 1-8: Person A writes, others contribute verbally. Focus: method (but every step must be justified by the theory — "We heat to 40°C because…").
• Minutes 9-16: Person B writes, others contribute. Focus: predicted results and theory explanation (linking the graph shape to collision theory).
• Minutes 17-25: Person C writes, others contribute. Focus: risk assessment and variables (linking risks to specific method steps, explaining controlled variables).
• Person D is the "quality checker" throughout — their job is to ensure every section connects to every other section.

Roles rotate, but the KEY design feature is that the person writing must incorporate the group's verbal input, not work alone. The other three are actively contributing ideas, not watching.

AI moves: See AI Facilitation Moves table below.

Phase 4: Presentation Preparation (10 minutes) The AI randomly selects one group member to present the report and answer the teacher's questions. The group has 10 minutes to ensure the selected person can explain EVERYTHING — not just "their" section.

AI move: At the start of Phase 4, the AI sends: "For the presentation, [randomly selected name] will explain and answer questions on the ENTIRE report. Group: you have 10 minutes to make sure [name] can explain the method, the theory, the predicted results, and the risk assessment. Quiz them. Find the gaps."

Phase 5: Presentation (5 minutes per group) Selected member presents. Teacher asks questions. This is the individual accountability mechanism — every member must understand the whole report because any of them could be selected.

AI Facilitation Moves

Trigger	AI Move	Purpose	Example
End of Phase 1 (5 min)	Round-robin prompt	Ensure all voices are heard before discussion begins	"Each person shares ONE thing from their individual notes. Go clockwise. Everyone speaks once before anyone speaks twice."
Phase 2, minute 10	Comprehension check	Verify shared understanding before writing begins	"Can everyone explain WHY higher temperature increases rate? Not just 'what happens' but 'why it happens.' Check with each other."
Phase 3, minute 5	Integration prompt	Prevent method from being written without theory connection	"I notice you're writing method steps. For each step, add: 'We chose this because...' linking it to your theory understanding."
Phase 3, one student not contributing for 3+ min	Participation prompt	Prevent disengagement	"[Name], what do you think about [the current section being written]? The group needs your input here."
Phase 3, minute 20	Progress check	Ensure the group is on track to finish	"You have 5 minutes left for writing. Quick check: have you connected the method to the theory, and the predicted graph to the collision model? What's still missing?"
Start of Phase 4	Random selection + preparation prompt	Create individual accountability	"[Name] has been selected to present. Group: 10 minutes to make sure they can explain everything."

Regulation Scaffolds

Planning (start of Phase 2): The AI sends: "Before you dive in — take 60 seconds as a group to agree: what does a GOOD experiment design look like for this task? What will make yours better than average? Write down your group's success criteria."

Monitoring (mid-Phase 3): The AI sends: "Halfway through writing. Look back at the success criteria you set. Are you meeting them? If not, what do you need to adjust?"

Reflection (end of Phase 3, before presentation prep): The AI sends: "Before you prepare for the presentation — one quick reflection as a group: what was the hardest part of working together on this? What would you do differently next time?" This builds metacognitive awareness of the collaboration process itself.

Equity Mechanisms

1. Individual thinking before group discussion (Phase 1). Every student arrives at the discussion with their own ideas. This prevents the loudest student from dominating from the start — everyone has something prepared.

2. Round-robin sharing (start of Phase 2). The AI enforces "everyone speaks once before anyone speaks twice." This ensures quiet students contribute before dominant students take over.

3. Rotating author structure (Phase 3). No single student can write the entire report. The rotation forces different people to hold the pen, reducing the "one person does everything" problem.

4. Quality checker role. Person D's explicit job is to ensure integration — but also to notice if anyone has stopped contributing and to pull them back in.

5. Random presentation selection (Phase 4). Because any member could be selected, the group is incentivised to ensure everyone understands everything. The free-rider who "did nothing" becomes a liability the group will actively address.

6. AI participation monitoring. The AI detects extended silence from individual students and prompts the group to include them. This catches disengagement that the teacher would miss.

What the AI Does NOT Do

• The AI does NOT answer chemistry questions. If a student asks the AI "What is collision theory?" the AI should respond: "That's a great question for your group. Ask them — and if nobody knows, flag it for your teacher."
• The AI does NOT write any part of the report. It does not generate text, suggest method steps, or draw predicted graphs.
• The AI does NOT evaluate the quality of the science. It doesn't say "your method is good" or "your theory explanation is incomplete." Quality evaluation is the teacher's role.
• The AI does NOT resolve disagreements. If the group disagrees about the predicted graph shape, the AI should say: "You have a disagreement. Good — that means someone is thinking deeply. Can each person explain their reasoning? Then decide as a group." The AI facilitates the PROCESS of disagreement resolution, not the CONTENT.

---

Known Limitations

1. AI-facilitated collaboration depends on reliable AI monitoring. The facilitation moves above assume the AI can detect participation imbalances, silence, and off-task behaviour. Current AI systems have variable capabilities here — text-based monitoring is feasible, but detecting disengagement in a face-to-face group working on paper is much harder. The design may need to be adapted for the specific AI system available.

2. The evidence for AI-specific facilitation of collaboration is still emerging. Dillenbourg (1999), Slavin (1995), and Järvelä & Hadwin (2013) established the principles of effective collaboration, but the specific application of AI to facilitate these principles is a newer field with fewer controlled studies. The AI facilitation moves above are principled extrapolations from established research, not empirically validated AI interventions.

3. Collaborative cognitive load can be high. Kirschner et al. (2018) note that coordination costs (communicating, negotiating, managing different perspectives) consume cognitive resources. For students who are already struggling with the content, the additional load of structured collaboration may be overwhelming. The teacher should monitor whether the collaboration structure is helping or hindering learning.

4. Cultural norms affect collaboration. Dillenbourg's (1999) framework was developed primarily in Western educational contexts. In some cultural contexts, direct disagreement, challenging peers' ideas, or speaking before someone of higher perceived status may be uncomfortable or inappropriate. The AI facilitation prompts may need cultural adaptation.

AI Feedback Design Principles

What This Skill Does

Evaluates a proposed AI feedback design against research criteria for effective automated feedback and suggests specific improvements. This skill takes a feedback scenario (what the student did) and the current or proposed AI response (what the system says), then analyses the feedback against principles from Shute (2008), Narciss (2008), Hattie & Timperley (2007), and emerging LLM feedback research (Dai et al., 2023). The output includes a diagnosis of what's working and what isn't, a redesigned version of the feedback, and practical implementation guidance. The core challenge is that most AI feedback falls into one of two failure modes: it's either too vague to be actionable ("Good effort! Try to improve your argument.") or too specific and does the thinking for the student ("Your thesis should be: Climate change is the defining challenge of our generation because…"). Effective feedback lives in the narrow space between these extremes — specific enough that the student knows what to do, but not so specific that it bypasses the cognitive work that produces learning. AI is specifically valuable here because it can generate feedback at scale, but this makes the design principles even more critical: bad feedback at scale is worse than no feedback at all.

Evidence Foundation

Hattie & Timperley (2007) conducted a meta-analysis finding that feedback has an average effect size of 0.73 — making it one of the most powerful influences on learning. However, they found enormous variation: some feedback interventions produced large positive effects while others had zero or even NEGATIVE effects. The critical variable was not whether feedback was given, but WHAT KIND of feedback was given. They proposed a model with four levels: task feedback (is the answer correct?), process feedback (what strategies can improve the work?), self-regulation feedback (how can you monitor your own learning?), and self feedback (you're a great student!). Task and process feedback were most effective; self feedback ("Good job!") was least effective and sometimes harmful because it directs attention to the self rather than the task. Shute (2008) reviewed formative feedback research and identified key principles: effective feedback is specific, timely, non-threatening, and focused on the task rather than the learner. She distinguished between verification feedback (correct/incorrect), elaborated feedback (why it's correct/incorrect and what to do next), and various combinations. She found that elaborated feedback generally outperforms simple verification, BUT that overly detailed feedback can overwhelm novice learners — creating a feedback paradox where more information sometimes produces less learning. Narciss (2008) developed the Informative Tutoring Feedback (ITF) model, which specifies that effective feedback should include: knowledge of result (correct or not), knowledge of the correct response (if wrong), and elaboration on the error (why it's wrong and what misconception it reveals). Critically, Narciss found that the optimal feedback depends on the error type: conceptual errors benefit from elaborated feedback, while careless slips benefit from simple verification. Kluger & DeNisi (1996) found in their meta-analysis that feedback that directs attention to the self (rather than the task) can DECREASE performance — a finding with direct implications for AI systems that generate encouraging but empty praise. Dai et al. (2023) evaluated LLM-generated feedback and found that while LLMs can produce fluent, well-structured feedback, they tend toward a specific pattern: excessive positivity, vague suggestions, and a reluctance to identify specific errors — precisely the pattern that research identifies as least effective.

Input Schema

The teacher must provide:

• Feedback scenario: What the student did. e.g. "Year 9 student submitted a persuasive essay arguing that school uniforms should be abolished. The argument is passionate but relies entirely on personal anecdotes — no evidence, no counterargument addressed, weak logical structure" / "Year 7 student solved 3x + 5 = 20 and got x = 7 (incorrect — should be x = 5)" / "A-level student wrote a lab report with correct data but a conclusion that doesn't follow from the results"
• Current feedback design: What the AI currently says or plans to say. e.g. "Great essay! You clearly feel strongly about this topic. To improve, try adding some evidence and considering the other side of the argument" / "Incorrect. The answer is x = 5. Try again" / "Your conclusion needs work. Think about what your data actually shows"

Optional (injected by context engine if available):

• Student level: Year group and proficiency
• Subject area: The curriculum subject
• Feedback goals: What the feedback should achieve
• System constraints: Technical or practical limitations

Prompt

You are an expert in the science of feedback in learning, with deep knowledge of Hattie & Timperley's (2007) feedback model (task, process, self-regulation, self levels), Shute's (2008) formative feedback principles, Narciss's (2008) Informative Tutoring Feedback model, Kluger & DeNisi's (1996) meta-analysis on feedback interventions, and emerging research on LLM-generated feedback quality (Dai et al., 2023). You understand that feedback is one of the most powerful influences on learning — and one of the most dangerous when poorly designed. You know that AI systems tend toward a specific failure mode: generating feedback that is positive, fluent, well-structured, and educationally useless.

CRITICAL PRINCIPLES:
- **Feedback must be SPECIFIC and ACTIONABLE.** "Good effort" is not feedback. "Your introduction states a position but doesn't preview your three supporting arguments — add a sentence that maps out your essay structure" IS feedback. If a student cannot read the feedback and know EXACTLY what to do next, it has failed.
- **Distinguish verification, elaboration, and strategic feedback.** Verification: "This is incorrect." Elaboration: "This is incorrect because you subtracted 5 from the left side but not the right." Strategic: "When you get stuck on equations, always check: did I do the same operation to both sides?" Different errors need different types. A conceptual error needs elaboration. A careless slip needs verification. A recurring pattern needs strategic feedback.
- **Avoid the positivity trap.** AI systems default to excessive positivity. "Great work!" before pointing out fundamental errors sends a contradictory signal and dilutes the corrective message. Positive feedback is appropriate ONLY when genuinely earned AND directed at specific features ("Your use of statistical evidence in paragraph 2 is effective because it directly supports your claim"). Generic praise is worse than no praise at all (Kluger & DeNisi, 1996).
- **Don't do the student's thinking.** Feedback that tells the student exactly what to write, what the answer is, or how to fix their work is not feedback — it's answer-giving. The goal is to close the gap between current and desired performance by showing the student WHERE the gap is and giving them enough information to close it themselves.
- **Match feedback complexity to student level.** Novice learners benefit from simple, clear feedback focused on one or two specific issues. Advanced learners benefit from more complex feedback that addresses multiple dimensions. Overloading novices with comprehensive feedback produces cognitive overload, not learning (Shute, 2008).

Your task is to evaluate and improve this feedback design:

**Feedback scenario:** {{feedback_scenario}}
**Current feedback design:** {{current_feedback_design}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Student level:** {{student_level}} — if not provided, infer from the scenario.
**Subject area:** {{subject_area}} — if not provided, infer from the scenario.
**Feedback goals:** {{feedback_goals}} — if not provided, assume the goal is to help the student improve their work while preserving their ownership of the thinking.
**System constraints:** {{system_constraints}} — if not provided, assume no significant constraints.

Return your output in this exact format:

## Feedback Evaluation: [Brief Scenario Description]

**Scenario:** [What the student did]
**Current feedback:** [What the AI currently says]
**Verdict:** [One-sentence summary — is this feedback likely to improve learning, have no effect, or actively harm it?]

### Diagnosis

[Analyse the current feedback against each principle. What works? What doesn't? Be specific — quote the problematic parts of the feedback and explain WHY they are problematic, citing the relevant research.]

### Feedback Type Analysis

| Feedback Component | Type | Effectiveness | Issue |
|---|---|---|---|
| [Quote from current feedback] | [Verification / Elaboration / Strategic / Self] | [Effective / Ineffective / Harmful] | [Why] |

### Improved Feedback Design

[The redesigned feedback. Show the exact text the AI should present to the student. Include specific, actionable guidance that addresses the identified weaknesses without doing the student's thinking for them.]

**Redesigned feedback:**

> [The exact feedback text]

### Design Rationale

[Explain why the improved version is better — what principles it follows and what specific changes were made.]

### Implementation Notes

[Practical guidance for deploying this feedback pattern — when it should trigger, how to handle edge cases, and what to watch for.]

**Self-check before returning output:** Verify that (a) the improved feedback is specific and actionable, (b) it doesn't do the student's thinking, (c) it avoids empty praise, (d) it uses the right type of feedback for the error type, and (e) it's appropriate for the student's level.

Example Output

Scenario: Feedback scenario: "Year 9 student submitted a persuasive essay arguing that school uniforms should be abolished. The argument is passionate but relies entirely on personal anecdotes — no evidence, no counterargument addressed, weak logical structure. The essay is about 400 words, has an introduction, three body paragraphs, and a conclusion." / Current feedback design: "Great essay! You clearly feel strongly about this topic and your writing voice is engaging. To improve, try adding some evidence to support your points and consider addressing the opposing viewpoint. Your conclusion could be stronger — try restating your main argument. Overall, a solid effort — keep it up! 7/10" / Student level: "Year 9, target grade 6 (UK GCSE equivalent)" / Subject area: "English Language — persuasive writing"

---

Feedback Evaluation: Year 9 Persuasive Essay — School Uniforms

Scenario: Year 9 student wrote a persuasive essay relying on personal anecdotes with no evidence, no counterargument, and weak structure. Current feedback: "Great essay! You clearly feel strongly… try adding some evidence… consider addressing the opposing viewpoint… a solid effort — keep it up! 7/10" Verdict: This feedback is unlikely to produce improvement. It is vague, excessively positive, and tells the student they're doing well when their essay has fundamental structural problems.

Diagnosis

Problem 1: The positivity trap. The feedback opens with "Great essay!" — but the essay has fundamental weaknesses (no evidence, no counterargument, weak structure). This sends a contradictory signal: you're telling the student the essay is great while simultaneously listing everything that's wrong with it. Kluger & DeNisi (1996) found that feedback directing attention to the self ("you clearly feel strongly") rather than the task reduces learning. The student's takeaway: "I got 7/10 and the teacher liked my voice. I just need to tweak a few things."

Problem 2: Vague improvement suggestions. "Try adding some evidence" — what kind of evidence? Where in the essay? How much? "Consider addressing the opposing viewpoint" — which opposing viewpoint? Where in the essay should it go? How should it be structured? Shute (2008) found that feedback must be specific enough that the student knows exactly what to do. These suggestions are too vague to be actionable.

Problem 3: No diagnosis of the actual problem. The essay's fundamental issue is that the student doesn't understand the difference between assertion and argument. Every paragraph makes claims supported only by personal experience. The feedback doesn't name this problem — it treats the symptoms (no evidence, no counterargument) without identifying the underlying cause (the student equates strong feeling with strong argument).

Problem 4: The score undermines the feedback. 7/10 signals "this is good." If the essay has no evidence and no counterargument, it has not met the basic criteria for persuasive writing. The score and the feedback are in conflict. The student will anchor on the score and dismiss the suggestions.

Feedback Type Analysis

Feedback Component	Type	Effectiveness	Issue
"Great essay!"	Self	Harmful	Empty praise directing attention to self not task
"You clearly feel strongly"	Self	Ineffective	Describes the student, not the work
"Your writing voice is engaging"	Task	Mildly effective	Specific enough to be useful, but buried in generic praise
"Try adding some evidence"	Strategic	Ineffective	Too vague — what evidence, where, how?
"Consider addressing the opposing viewpoint"	Strategic	Ineffective	Too vague — which viewpoint, where in the essay?
"Your conclusion could be stronger"	Task	Ineffective	"Stronger" is meaningless without specifics
"A solid effort — keep it up!"	Self	Harmful	False reassurance when major revisions needed
"7/10"	Verification	Harmful	Score contradicts the need for fundamental revision

Improved Feedback Design

Redesigned feedback:

What's working: Your writing voice is confident and direct — "I shouldn't have to wear someone else's idea of who I am every single day" is a strong, memorable line. Hold onto that voice.

The main issue — assertion vs. argument: Right now, every paragraph makes a claim and supports it with your personal experience. That's assertion, not argument. Persuasive writing needs EVIDENCE — facts, statistics, expert opinions, or real examples that someone who disagrees with you would have to take seriously. Your reader is thinking: "That's how YOU feel, but why should I care?"

One specific fix to try first: Pick your strongest paragraph (I'd suggest paragraph 2 about comfort). Find ONE piece of evidence that supports your point — a study about student concentration, a school that removed uniforms and measured the effect, or a quote from an education expert. Add it. Then look at what it does to the paragraph. That's the difference between assertion and argument.

Next step after that: Choose the strongest argument AGAINST your position (probably cost — uniforms save families money). Write a paragraph that acknowledges this argument and then explains why your position is still stronger. This is called a counterargument, and it's the single most persuasive move in argumentative writing — it shows you've considered the other side and still have a stronger case.

Don't revise yet. First, reread your essay and put a star next to every sentence that is supported by evidence (not personal experience). Count the stars. That count tells you how much work is needed.

Design Rationale

The redesigned feedback makes five key changes:

1. Specific positive feedback first, directed at the task. "Your writing voice is confident and direct" with a quoted example — this is task-level feedback on a genuine strength, not generic self-level praise. It tells the student exactly what to keep doing.

2. Names the underlying problem. "Assertion vs. argument" gives the student a conceptual framework for understanding what's wrong. This is elaborated feedback (Narciss, 2008) — it doesn't just say "add evidence" but explains WHY evidence is needed and what the essay currently lacks.

3. One specific, manageable action. Instead of a list of vague improvements, the feedback gives ONE concrete task: find one piece of evidence for one paragraph. This follows Shute's (2008) principle of matching feedback complexity to learner level — a Year 9 student targeting grade 6 needs focused, achievable steps, not a comprehensive critique.

4. The counterargument instruction is scaffolded. It doesn't just say "address the opposing viewpoint" — it tells the student which opposing argument to choose and explains the strategic reason for including it. This is process-level feedback (Hattie & Timperley, 2007).

5. The diagnostic task replaces the score. "Count the stars" is a self-regulation prompt that helps the student see the problem for themselves, rather than being told. This develops metacognitive awareness — a higher-order feedback function that builds independence.

Implementation Notes

• Do not open with a score. If a score is required by the system, place it at the end, after the actionable feedback. Better yet, withhold the score until after revision.
• Limit feedback to 2-3 actionable points. This example addresses two issues (evidence, counterargument) plus one metacognitive prompt. That's the maximum for a Year 9 student. More points produce cognitive overload and paralysis.
• Watch for the LLM positivity pattern. If the AI is generating feedback, audit it for the "Great job! But…" pattern (Dai et al., 2023). Instruct the model to lead with specific, earned praise or to lead with the diagnostic — not with generic enthusiasm.
• Re-evaluate after revision. The feedback above is designed to produce a specific revision (adding evidence to one paragraph, adding a counterargument paragraph). After revision, the feedback should address different issues — probably structure and logical flow. Don't repeat the same feedback if the student has acted on it.

---

Known Limitations

1. This skill evaluates feedback DESIGN, not feedback DELIVERY. The same feedback text can be effective or harmful depending on timing, student emotional state, and the relationship between student and system. A student who has just failed three assignments in a row needs a different emotional register than a confident student who made a careless error. This skill addresses content and structure, not affect and timing.

2. The LLM feedback evidence base is still emerging. Dai et al. (2023) is one of the first large-scale studies of LLM feedback quality, and the field is developing rapidly. The principles from Shute (2008), Narciss (2008), and Hattie & Timperley (2007) are well-established for human feedback — their application to AI-generated feedback is theoretically sound but not yet comprehensively validated.

3. Cultural context affects feedback norms. The direct, task-focused feedback style recommended here reflects Western educational research norms. In some cultural contexts, direct criticism (even when constructive) may be received differently. Narciss's (2008) model was developed primarily in European and North American contexts.

4. Feedback interacts with student self-efficacy in complex ways. Kluger & DeNisi (1996) found that feedback can decrease performance when it threatens self-concept. For students with very low self-efficacy, the "no empty praise" principle needs to be balanced against the risk of further damaging motivation. This skill does not model the individual student's motivational state.

Cognitive Tutoring Architecture Designer

What This Skill Does

Designs the cognitive architecture for a tutoring system — the knowledge component model, knowledge tracing algorithm, problem selection logic, and feedback architecture that together create an adaptive learning experience. This skill is based on the Cognitive Tutor architecture developed by Anderson et al. (1995) at Carnegie Mellon University, which remains one of the most empirically validated approaches to intelligent tutoring. The Cognitive Tutor approach is built on ACT-R theory: the idea that expertise in any domain can be decomposed into a set of discrete "knowledge components" (production rules) that can be individually learned, practised, and tracked. The system maintains a model of what each student knows (knowledge tracing), selects problems that target the knowledge components the student has not yet mastered, and provides step-level feedback when errors occur. Pane et al. (2014) conducted a large-scale RAND evaluation finding that Cognitive Tutor Algebra I improved student outcomes compared to traditional instruction. AI is specifically valuable here because the architecture requires real-time tracking of individual knowledge states across many components — a task that is computationally straightforward but humanly impossible for a teacher managing 30 students.

Evidence Foundation

Anderson et al. (1995) developed the Cognitive Tutor based on ACT-R (Adaptive Control of Thought—Rational), a cognitive architecture that models human learning as the acquisition of production rules — IF-THEN knowledge components. For example, in algebra, one knowledge component might be "IF the equation has the form ax + b = c, THEN subtract b from both sides as the first step." The key insight: complex skills can be decomposed into a finite set of these components, each of which can be independently tracked and tutored. The Cognitive Tutor maintained a student model that estimated the probability that each student had learned each knowledge component, and selected problems that targeted the weakest components. Corbett & Anderson (1995) developed knowledge tracing (KT), the algorithm that tracks student mastery. KT uses a hidden Markov model: each knowledge component is either "learned" or "unlearned" (the hidden state). Each time a student attempts a problem step that involves a knowledge component, the system observes whether the step was correct or incorrect (the observable) and updates its estimate of the probability that the component is learned. Four parameters govern the model: P(L₀) — the probability the student knew the component before instruction; P(T) — the probability of learning the component at each opportunity; P(G) — the probability of guessing correctly without knowing; and P(S) — the probability of making a slip despite knowing. When P(Lₙ) exceeds a threshold (typically 0.95), the component is considered mastered. Koedinger & Aleven (2007) articulated the "assistance dilemma" in the context of cognitive tutors: providing too much assistance (immediate hints, step-level guidance) can lead students to learn the tutor's interface rather than the domain knowledge. They found that the optimal level of assistance depended on the student's current knowledge state — struggling students benefited from more assistance, while advanced students benefited from less. Ritter et al. (2007) reported on the deployment of Cognitive Tutor Algebra in over 2,600 schools, demonstrating the approach's scalability. Pane et al. (2014) conducted a RAND-funded randomised controlled trial of Cognitive Tutor Algebra I and found modest but significant positive effects on algebra achievement, particularly for students who used the system as designed (with fidelity to the intended model).

Input Schema

The teacher must provide:

• Skill domain: What students need to learn. e.g. "Solving linear equations — from one-step (x + 5 = 12) through two-step (3x + 5 = 20) to multi-step equations with variables on both sides (4x + 3 = 2x + 11)" / "Balancing chemical equations" / "Constructing and interpreting histograms from raw data"
• Knowledge components: The building blocks. e.g. "For linear equations: (1) identifying the unknown, (2) inverse operations (addition ↔ subtraction), (3) inverse operations (multiplication ↔ division), (4) order of operations (undo addition/subtraction before multiplication/division), (5) applying operation to both sides, (6) combining like terms, (7) moving variables to one side" / "For histograms: (1) identifying continuous data, (2) choosing appropriate class intervals, (3) calculating frequency for each interval, (4) scaling axes, (5) drawing bars with no gaps, (6) reading values from a histogram, (7) comparing distributions"

Optional (injected by context engine if available):

• Student level: Year group and proficiency
• Subject area: The curriculum subject
• Common errors: Frequent mistakes and their causes
• Mastery threshold: What counts as mastered
• System context: AI, teacher-delivered, or hybrid

Prompt

You are an expert in cognitive tutoring architecture, with deep knowledge of Anderson et al.'s (1995) ACT-R-based Cognitive Tutor framework, Corbett & Anderson's (1995) knowledge tracing algorithm, Koedinger & Aleven's (2007) assistance dilemma research, Ritter et al.'s (2007) large-scale deployment data, and Pane et al.'s (2014) RAND evaluation. You understand that effective tutoring systems are built on a precise cognitive model — a decomposition of the target skill into knowledge components that can be individually tracked, practised, and mastered. You know that the quality of the cognitive model determines the quality of the tutoring: a poorly decomposed skill set leads to misdiagnosis, inappropriate problem selection, and ineffective feedback.

CRITICAL PRINCIPLES:
- **Decompose into TESTABLE knowledge components.** Each component must be independently observable — there must be a problem step where the component is required and where success or failure reveals whether the student has learned it. If you can't design a problem step that isolates a component, the decomposition is too coarse.
- **Knowledge components should be at the GRAIN SIZE of learning.** Too coarse (e.g., "can solve equations") and you can't diagnose specific weaknesses. Too fine (e.g., "can subtract 7 from 22") and you're tracking arithmetic, not algebra. The right grain size is the level at which students make conceptually meaningful errors.
- **Map the dependency structure.** Some components depend on others — you can't apply an operation to both sides if you don't know the inverse operation. The problem selection logic should respect these dependencies: don't present problems requiring Component 5 if the student hasn't mastered Components 1-4.
- **Link errors to components.** Each common error should be traceable to a specific knowledge component that has not been mastered. If a student subtracts from only one side, that's a failure of the "apply to both sides" component, not a general "equation" failure. The feedback should target the specific component.
- **Mastery is probabilistic, not binary.** Knowledge tracing gives a PROBABILITY of mastery, not a certainty. A student who gets 3 in a row right might still be guessing (P(G) > 0). A student who gets one wrong might have slipped (P(S) > 0). The system should require consistent performance before declaring mastery.

Your task is to design a cognitive tutoring architecture for:

**Skill domain:** {{skill_domain}}
**Knowledge components:** {{knowledge_components}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Student level:** {{student_level}} — if not provided, design for a general secondary school context.
**Subject area:** {{subject_area}} — if not provided, infer from the domain.
**Common errors:** {{common_errors}} — if not provided, identify the most likely errors based on the knowledge components.
**Mastery threshold:** {{mastery_threshold}} — if not provided, use P(Lₙ) ≥ 0.95 (approximately 3-4 consecutive correct applications).
**System context:** {{system_context}} — if not provided, design for an AI tutoring system.

Return your output in this exact format:

## Cognitive Tutoring Architecture: [Skill Domain]

**Domain:** [What students are learning]
**Knowledge components:** [How many, at what grain size]
**Architecture basis:** [ACT-R / knowledge tracing — cite the specific principles used]

### Knowledge Component Map

[Complete map of all knowledge components, their descriptions, their dependencies, and the observable evidence of mastery for each]

| KC ID | Knowledge Component | Description | Depends On | Evidence of Mastery | Common Error |
|---|---|---|---|---|---|
| [ID] | [Name] | [What the student knows/can do] | [Prerequisites] | [What correct performance looks like] | [What incorrect performance looks like] |

### Dependency Graph

[Visual or textual representation of which components must be mastered before others can be addressed]

### Knowledge Tracing Design

[How the system tracks mastery — the algorithm, parameters, and mastery threshold]

### Problem Selection Logic

[How the system chooses the next problem — which components to target, how to combine components, and how difficulty increases]

### Feedback Architecture

[What feedback to provide for each error type — linked to specific knowledge components]

### Example Tutoring Sequence

[A realistic sequence showing how the system would tutor a student through 5-8 problems, adapting based on their responses]

**Self-check before returning output:** Verify that (a) each knowledge component is independently testable, (b) components are at the right grain size, (c) dependencies are mapped, (d) errors are linked to specific components, and (e) mastery is treated probabilistically.

Example Output

Scenario: Skill domain: "Solving linear equations — from one-step (x + 5 = 12) through two-step (3x + 5 = 20) to equations with variables on both sides (4x + 3 = 2x + 11)" / Knowledge components: "(1) identifying the unknown, (2) inverse of addition/subtraction, (3) inverse of multiplication/division, (4) order of undoing operations, (5) applying operation to both sides, (6) combining like terms, (7) collecting variables to one side" / Student level: "Year 8, first term of formal algebra" / Subject area: "Mathematics" / Common errors: "Subtracting from one side only, wrong operation order (dividing before subtracting), not understanding that 3x means 3 × x, combining unlike terms (adding 3x + 5 to get 8x)"

---

Cognitive Tutoring Architecture: Solving Linear Equations

Domain: Linear equations from one-step to variables on both sides Knowledge components: 7, at the conceptual-procedural grain size Architecture basis: ACT-R production rule model (Anderson et al., 1995) with Bayesian knowledge tracing (Corbett & Anderson, 1995)

Knowledge Component Map

KC ID	Knowledge Component	Description	Depends On	Evidence of Mastery	Common Error
KC1	Identify the unknown	Recognise that x (or any letter) represents an unknown number that the equation will reveal	None	Student can state "We need to find the value of x"	Student treats x as a label rather than an unknown number; attempts to "solve" by inspection without any algebraic steps
KC2	Inverse of addition/subtraction	Knows that the inverse of +b is −b and vice versa	KC1	Given x + 5 = 12, student identifies "subtract 5" as the correct operation	Student adds instead of subtracting (or vice versa); adds 5 to 12 instead of subtracting
KC3	Inverse of multiplication/division	Knows that the inverse of ×a is ÷a and vice versa; understands that 3x means 3 × x	KC1	Given 3x = 15, student identifies "divide by 3" as the correct operation	Student subtracts 3 instead of dividing; doesn't understand that 3x = 3 × x
KC4	Operation order	When undoing operations, reverse the order of BIDMAS — undo addition/subtraction BEFORE multiplication/division	KC2, KC3	Given 3x + 5 = 20, student subtracts 5 first, then divides by 3	Student divides by 3 first, producing (3x + 5)/3 = 20/3 — correct algebra but much harder to simplify, or produces an error by dividing only the 3x by 3
KC5	Both sides	Any operation performed on one side must be performed on the other side	KC2, KC3	Student writes "3x + 5 − 5 = 20 − 5" (showing subtraction on BOTH sides)	Student writes "3x = 20 − 5" without subtracting from the left, or subtracts 5 from the left but not the right
KC6	Combining like terms	Can simplify expressions by combining terms with the same variable (3x + 2x = 5x) and constant terms (7 + 3 = 10), and knows NOT to combine unlike terms	KC1	Given 3x + 2x + 5 = 20, student correctly simplifies to 5x + 5 = 20	Student combines unlike terms: 3x + 5 = 8x; or fails to combine like terms when they should
KC7	Collecting variables	Can move variable terms to one side by applying inverse operations to variable terms	KC2, KC5	Given 4x + 3 = 2x + 11, student subtracts 2x from both sides to get 2x + 3 = 11	Student subtracts numbers when they should subtract variables; or tries to subtract 4x and 2x simultaneously from the same side

Dependency Graph

KC1 (Identify unknown)
├── KC2 (Inverse +/−)
│   ├── KC4 (Operation order) ← also requires KC3
│   └── KC5 (Both sides) ← also requires KC3
├── KC3 (Inverse ×/÷)
│   ├── KC4 (Operation order) ← also requires KC2
│   └── KC5 (Both sides) ← also requires KC2
└── KC6 (Combining like terms)

KC5 (Both sides) + KC2 → KC7 (Collecting variables)
KC4 + KC5 + KC6 → Multi-step equation competence
KC4 + KC5 + KC6 + KC7 → Variables-on-both-sides competence

Knowledge Tracing Design

Algorithm: Bayesian Knowledge Tracing (Corbett & Anderson, 1995)

Parameters (initial values, calibrated per domain):

Parameter	Value	Meaning
P(L₀)	0.10	Probability student knows the component before any instruction (low for novel algebra concepts)
P(T)	0.20	Probability of learning the component at each practice opportunity
P(G)	0.15	Probability of guessing correctly without knowing (low for multi-step algebra)
P(S)	0.10	Probability of slipping (making an error despite knowing)
Mastery threshold	0.95	P(Lₙ) at which the component is considered mastered

Update rule: After each student response on a problem step involving KC_i:

• If correct: P(Lₙ) = P(Lₙ₋₁)(1 - P(S)) / [P(Lₙ₋₁)(1 - P(S)) + (1 - P(Lₙ₋₁))P(G)] — then apply learning: P(Lₙ) = P(Lₙ) + (1 - P(Lₙ))P(T)
• If incorrect: P(Lₙ) = P(Lₙ₋₁)P(S) / [P(Lₙ₋₁)P(S) + (1 - P(Lₙ₋₁))(1 - P(G))] — then apply learning: P(Lₙ) = P(Lₙ) + (1 - P(Lₙ))P(T)

In plain English: Each time a student attempts a step, the system updates its belief about whether they know the relevant knowledge component. A correct answer increases the mastery estimate; an incorrect answer decreases it. But the system accounts for lucky guesses (P(G)) and careless slips (P(S)), so one correct answer doesn't prove mastery and one error doesn't prove ignorance. Approximately 3-4 consecutive correct applications are needed to reach the 0.95 mastery threshold.

Problem Selection Logic

Rule 1: Respect dependencies. Never present a problem requiring KC_n if the student has not mastered all prerequisite KCs.

Rule 2: Target the weakest unmastered component. Select problems that involve the knowledge component with the lowest P(Lₙ) among those whose prerequisites are met.

Rule 3: Gradual component combination. Start with problems requiring only 1-2 KCs. As components are mastered, introduce problems requiring 3-4 KCs simultaneously. This controls cognitive load.

Rule 4: Interleave mastered components. Occasionally include a step requiring a previously mastered KC to prevent forgetting and to verify that mastery is retained.

Problem difficulty progression:

Level	KCs Required	Example Problem	When to Present
1	KC1, KC2	x + 5 = 12	Start
2	KC1, KC3	3x = 15	KC2 mastered
3	KC1, KC2, KC3, KC4, KC5	3x + 5 = 20	KC2, KC3 mastered
4	KC1-KC6	3x + 2x + 5 = 20	KC4, KC5 mastered
5	KC1-KC7	4x + 3 = 2x + 11	KC4, KC5, KC6 mastered

Feedback Architecture

Error Pattern	Likely KC Failure	Feedback
Student attempts to guess x by inspection	KC1	"This equation needs algebraic steps, not guessing. Your goal is to get x by itself on one side. What operation would help?"
Student adds instead of subtracting (or vice versa)	KC2	"You added 5 to both sides. But x already HAS 5 added to it. What's the OPPOSITE of adding 5?"
Student subtracts 3 from 3x instead of dividing	KC3	"Remember: 3x means 3 TIMES x. To undo multiplication, you need to... ?"
Student divides before subtracting in 3x + 5 = 20	KC4	"You went straight to dividing by 3. But the +5 is in the way. Try undoing the addition first — that's the reverse of the order the expression was built."
Student operates on one side only	KC5	"You subtracted 5 from the left side. What about the right side? An equation is like a balance — what you do to one side, you must do to the other."
Student writes 3x + 5 = 8x	KC6	"You added 3x and 5 to get 8x. But 3x and 5 are different types of terms — one has an x and one doesn't. You can only combine terms that are the SAME type."
Student confused about moving variables	KC7	"You have x's on BOTH sides: 4x + 3 = 2x + 11. To solve, get all the x terms on ONE side first. How could you remove the 2x from the right side?"

Example Tutoring Sequence

A student begins with P(L₀) = 0.10 for all components.

Problem 1: x + 8 = 15 (Tests KC1, KC2, KC5)

• Student writes: x = 15 - 8 = 7 ✓
• KC1: P(L) → 0.28, KC2: P(L) → 0.28, KC5: P(L) → 0.28

Problem 2: x - 3 = 10 (Tests KC1, KC2, KC5 — inverse direction)

• Student writes: x = 10 + 3 = 13 ✓
• KC1: P(L) → 0.51, KC2: P(L) → 0.51, KC5: P(L) → 0.51

Problem 3: x + 12 = 5 (Tests KC1, KC2, KC5 — negative result)

• Student writes: x = 5 + 12 = 17 ✗ (KC2 failure — added instead of subtracting)
• System feedback: "You added 12 to 5. But the equation says x PLUS 12 equals 5. To undo the +12, you need to subtract, not add."
• KC2: P(L) → 0.33 (decreased), KC5: not updated (error was KC2, not KC5)

Problem 4: x + 12 = 5 (Same problem, retry)

• Student writes: x = 5 - 12 = -7 ✓
• KC2: P(L) → 0.53

Problem 5: 4x = 20 (Tests KC1, KC3 — multiplication/division, new KC)

• Student writes: x = 20 ÷ 4 = 5 ✓
• KC3: P(L) → 0.28

Problem 6: 7x = 21 (Tests KC1, KC3)

• Student writes: x = 21 - 7 = 14 ✗ (KC3 failure — subtracted instead of dividing)
• System feedback: "7x means 7 TIMES x. To undo multiplication by 7, you need to divide, not subtract."
• KC3: P(L) → 0.17

[Sequence continues: after KC2 and KC3 reach mastery (~0.95), system introduces two-step equations requiring KC4 and KC5...]

---

Known Limitations

1. Knowledge component decomposition is an art, not a science. Different researchers decompose the same domain differently. Anderson et al.'s ACT-R framework provides theoretical guidance, but the practical decisions about grain size, component boundaries, and dependency structure require domain expertise and iterative testing. The decomposition above is a principled starting point, not a validated cognitive model.

2. Standard knowledge tracing assumes binary knowledge states. Corbett & Anderson's (1995) model assumes each KC is either "learned" or "unlearned." In reality, knowledge exists on a continuum: a student might partially understand inverse operations, correctly applying them in simple contexts but failing in complex ones. More advanced models (e.g., Deep Knowledge Tracing using neural networks) address this, but they sacrifice interpretability.

3. The RAND evaluation showed modest effect sizes. Pane et al. (2014) found that Cognitive Tutor Algebra I improved outcomes, but the effect size was relatively small (0.22 in the second year of implementation). The practical significance depends on context: a small average effect may mask larger effects for specific student subgroups. Implementation fidelity was a major moderator.

4. Cognitive tutors work best for well-structured domains. Linear equations, chemical formula balancing, and other procedural domains decompose cleanly into knowledge components. Ill-structured domains (essay writing, historical analysis, creative tasks) resist this decomposition. The cognitive tutor architecture is powerful but domain-limited.

Digital Worked Example Sequence

What This Skill Does

Designs a worked example sequence optimised for digital delivery — incorporating self-explanation prompts, a systematic fading schedule, and interactivity design that exploits what digital environments offer beyond paper. The sequence moves students from studying complete worked examples (where every step is shown and explained) through faded examples (where progressively more steps are removed for the student to complete) to independent problem-solving. The critical insight from Renkl (2014) is that worked examples only produce learning when students actively process them — passive reading of worked examples is barely better than no examples at all. Digital delivery creates unique opportunities (self-explanation prompts at each step, immediate feedback on faded steps, adaptive pacing based on performance) and unique risks (split attention between screen elements, cognitive overload from multimedia, temptation to click through without thinking). The output includes the complete example sequence, embedded self-explanation prompts, a precise fading schedule, and digital design specifications. AI is specifically valuable here because designing an effective digital worked example sequence requires coordinating content design (the mathematical or procedural steps), cognitive design (fading schedule, self-explanation points), and interface design (how steps are revealed, where prompts appear, how feedback is given) — three design dimensions that must be aligned.

Evidence Foundation

Sweller, van Merriënboer & Paas (2019) updated cognitive load theory for contemporary digital learning contexts, identifying new sources of extraneous load specific to digital environments: transient information (content that appears and disappears), split attention between multiple screen areas, and redundancy in multimedia presentations. They emphasised that digital worked examples must manage these load sources through careful design. Renkl (2014) synthesised 25 years of worked example research into instructional principles: examples should be structured (steps clearly delineated), self-explanation should be prompted (not left to chance), fading should be systematic (one step at a time, starting with the most recently learned), and the transition to independent practice should be gradual. Atkinson et al. (2000) established foundational principles: worked examples are most effective for NOVICE learners (experts suffer the "expertise reversal effect" — examples become redundant and counterproductive), examples should alternate with practice problems rather than being presented in blocks, and the key mechanism is schema acquisition (building mental templates for problem types). Renkl, Atkinson & Große (2004) demonstrated that systematic fading — removing solution steps one at a time — was significantly more effective than abrupt transitions from full examples to full problems. Wylie & Chi (2014) showed that self-explanation prompts embedded within multimedia worked examples dramatically improved learning compared to worked examples without prompts, because they forced active processing of each step.

Input Schema

The teacher must provide:

• Skill to teach: The specific procedure. e.g. "Solving simultaneous equations by elimination" / "Constructing a balanced argument paragraph" / "Converting between units of measurement" / "Writing a recursive function in Python"
• Target platform: Where it will be delivered. e.g. "Google Slides with embedded quizzes" / "Custom web app with step-by-step reveal" / "Interactive PDF" / "Learning management system (Canvas/Moodle)"

Optional (injected by context engine if available):

• Student level: Year group and proficiency
• Subject area: Curriculum subject
• Sequence length: Number of examples
• Interactivity level: Platform capabilities
• Student data available: Whether the system adapts

Prompt

You are an expert in digital worked example design, with deep knowledge of Sweller et al.'s (2019) updated cognitive load theory for digital contexts, Renkl's (2014) instructional principles for example-based learning, Atkinson et al.'s (2000) foundational worked example research, Renkl, Atkinson & Große's (2004) systematic fading procedure, and Wylie & Chi's (2014) research on self-explanation in multimedia learning. You understand that digital worked examples are not PDFs of paper examples displayed on a screen — they are interactive learning tools that exploit digital affordances (step-by-step reveal, embedded prompts, immediate feedback, adaptive pacing) while managing digital risks (transient information, split attention, passive clicking).

CRITICAL PRINCIPLES:
- **Self-explanation at EVERY step.** The single most powerful design feature is a self-explanation prompt after each step: "Why did we do this?" or "What rule is being applied here?" Without self-explanation, students click through examples like a slideshow — seeing without processing. Renkl (2014) showed that prompted self-explanation is the mechanism that converts passive observation into active learning.
- **Systematic fading, not abrupt transition.** The sequence should remove ONE step at a time, starting with the LAST step (backward fading) or the step that students find easiest. Each faded step becomes a "completion problem" — the student must supply the missing step. The transition from full example to independent problem should be so gradual that students barely notice when they're doing it on their own.
- **Digital pacing: reveal steps one at a time.** Don't display the entire solution at once — this creates extraneous load and encourages scanning rather than studying. Reveal one step at a time, with the self-explanation prompt BEFORE the next step is shown. This forces students to process each step before moving on.
- **Immediate, targeted feedback on faded steps.** When a student fills in a faded step, provide immediate feedback: correct (with brief confirmation) or incorrect (with a hint addressing the specific error, not just "try again"). This is where digital delivery excels over paper.
- **Manage split attention.** The problem, the example steps, the self-explanation prompts, and the feedback must be spatially integrated — not spread across different areas of the screen. The student should never have to hold information from one screen area in working memory while looking at another area.

Your task is to design a digital worked example sequence for:

**Skill to teach:** {{skill_to_teach}}
**Target platform:** {{target_platform}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Student level:** {{student_level}} — if not provided, design for secondary school novice learners.
**Subject area:** {{subject_area}} — if not provided, infer from the skill.
**Sequence length:** {{sequence_length}} — if not provided, design 6 examples (2 full, 2 faded, 2 independent).
**Interactivity level:** {{interactivity_level}} — if not provided, design for step-by-step reveal with text input for faded steps.
**Student data available:** {{student_data_available}} — if not provided, design a fixed sequence with optional adaptation notes.

Return your output in this exact format:

## Digital Worked Example Sequence: [Skill]

**Skill:** [What students learn]
**Platform:** [Where it's delivered]
**Sequence structure:** [How many examples, from full to faded to independent]

### Sequence Overview

[The progression from fully worked to fully independent — what each example does]

### Example [N]: [Full / Faded-1 / Faded-2 / Independent]

**Problem:** [The problem to solve]
**Steps shown vs. faded:**
[Which steps are shown, which are faded (student completes), which have self-explanation prompts]

**Step-by-step design:**
For each step:
- **Step [N]:** [The step — shown or faded]
- **Self-explanation prompt:** [The question students answer before seeing the next step]
- **If faded — expected student input:** [What the student should enter]
- **Feedback on faded step:** [Immediate feedback — correct or incorrect with hint]

### Fading Schedule

[The precise schedule: which steps fade in which example, and why this order]

### Digital Design Specifications

**Pacing:** [How steps are revealed — one at a time, with forced pause at self-explanation prompts]
**Layout:** [How the screen is organised — problem, steps, prompts, feedback integrated to avoid split attention]
**Feedback timing:** [When and how feedback appears]
**Navigation:** [Can students go back? Can they skip ahead? Design choices and rationale]

### Adaptation Notes

[How the sequence could adapt based on student performance — if the platform supports it]

**Self-check before returning output:** Verify that (a) every step has a self-explanation prompt, (b) fading is systematic and gradual, (c) steps are revealed one at a time, (d) feedback on faded steps is immediate and specific, and (e) split attention is managed through integrated layout.

Example Output

Scenario: Skill to teach: "Solving simultaneous equations by elimination — e.g., 2x + 3y = 12 and 4x - 3y = 6. Students need to: identify which variable to eliminate, add or subtract equations, solve for one variable, substitute back to find the other." / Target platform: "Custom web app — can reveal steps one at a time, accept text input, provide immediate feedback" / Student level: "Year 10, have just learned to solve single-variable linear equations, first encounter with simultaneous equations" / Sequence length: "6 examples (2 full, 2 faded, 2 independent)"

---

Digital Worked Example Sequence: Simultaneous Equations by Elimination

Skill: Solving simultaneous equations by elimination Platform: Custom web app (step-by-step reveal, text input, immediate feedback) Sequence structure: 6 examples — E1-E2 (fully worked), E3-E4 (progressively faded), E5-E6 (independent with hints available)

Sequence Overview

Example	Type	Steps Shown	Steps Faded	Purpose
E1	Full worked example	All 5 steps	None	Students study and self-explain every step
E2	Full worked example (different numbers)	All 5 steps	None	Reinforces the procedure with variation
E3	Faded — Step 5 removed	Steps 1–4	Step 5 (final substitution)	Student completes the easiest step
E4	Faded — Steps 4 and 5 removed	Steps 1–3	Steps 4–5 (solve and substitute)	Student completes the last two steps
E5	Independent (hints available)	None shown	All steps	Student solves alone, hint sequence available
E6	Independent (no hints)	None shown	All steps	Full independent practice

Example 1: Full Worked Example

Problem: Solve simultaneously: 2x + 3y = 12 (Equation 1) and 4x - 3y = 6 (Equation 2)

Step 1 — Identify the elimination opportunity Display: "Look at the y terms: +3y and -3y. They are the SAME size but OPPOSITE signs. If we add the equations, the y terms will cancel out." Self-explanation prompt: "Why can we add these two equations together? What happens to the y terms when we add +3y and -3y?" [Student types response before next step appears] Model answer shown after student responds: "+3y + (-3y) = 0. The y terms cancel because they are additive inverses."

Step 2 — Add the equations Display: "(2x + 3y) + (4x - 3y) = 12 + 6 → 6x + 0y = 18 → 6x = 18" Self-explanation prompt: "We added the LEFT sides together and the RIGHT sides together. Why is this allowed? (Hint: both equations are true, so both sides are equal.)"

Step 3 — Solve for x Display: "6x = 18 → x = 18 ÷ 6 → x = 3" Self-explanation prompt: "We now have a single-variable equation — you already know how to solve these. What operation did we use?"

Step 4 — Substitute back Display: "Substitute x = 3 into Equation 1: 2(3) + 3y = 12 → 6 + 3y = 12 → 3y = 6 → y = 2" Self-explanation prompt: "Why did we substitute into Equation 1 and not Equation 2? (Actually, either would work. Can you verify by substituting into Equation 2?)"

Step 5 — Check the solution Display: "Check: Equation 1: 2(3) + 3(2) = 6 + 6 = 12 ✓. Equation 2: 4(3) - 3(2) = 12 - 6 = 6 ✓. Solution: x = 3, y = 2." Self-explanation prompt: "Why is it important to check in BOTH equations, not just one?"

Example 3: Faded — Step 5 Removed

Problem: Solve simultaneously: 3x + 2y = 16 and 5x - 2y = 8

Steps 1–4 are shown (with self-explanation prompts) following the same structure as E1.

Step 5 — FADED (student completes): Display: "We found x = 3 and y = 3.5. Now CHECK your solution in both equations." Student input field: [Two boxes: "Equation 1: 3() + 2() = " and "Equation 2: 5() - 2(___) = ___"] Expected input: Equation 1: 3(3) + 2(3.5) = 9 + 7 = 16 ✓. Equation 2: 5(3) - 2(3.5) = 15 - 7 = 8 ✓. Feedback if correct: "Both equations check out. You've verified the solution is correct." Feedback if incorrect: "Check your arithmetic. Substitute x = 3 and y = 3.5 into each equation carefully."

Example 4: Faded — Steps 4 and 5 Removed

Problem: Solve simultaneously: x + 4y = 14 and 3x - 4y = 2

Steps 1–3 are shown. After adding equations: 4x = 16, so x = 4.

Step 4 — FADED: Display: "We know x = 4. Substitute into Equation 1 to find y." Student input field: [y = ___] Expected input: y = 2.5 (from 4 + 4y = 14 → 4y = 10 → y = 2.5) Feedback if correct: "Correct! Now check your solution." Feedback if incorrect (e.g., y = 10): "You may have forgotten to divide. 4y = 10 means y = 10 ÷ 4. Try again."

Step 5 — FADED: [As in E3]

Fading Schedule

Example	Step 1 (Identify)	Step 2 (Add/Subtract)	Step 3 (Solve)	Step 4 (Substitute)	Step 5 (Check)
E1	Shown	Shown	Shown	Shown	Shown
E2	Shown	Shown	Shown	Shown	Shown
E3	Shown	Shown	Shown	Shown	FADED
E4	Shown	Shown	Shown	FADED	FADED
E5	FADED	FADED	FADED	FADED	FADED
E6	FADED	FADED	FADED	FADED	FADED

Fading logic: Steps fade from LAST to FIRST (backward fading). Step 5 (checking) fades first because it is procedurally the simplest. Step 1 (identifying the elimination opportunity) fades last because it requires the most strategic thinking. This means students are independently performing the EASIER steps first while still being scaffolded on the HARDER steps.

Digital Design Specifications

Pacing: Each step is revealed by clicking a "Next Step" button. The self-explanation prompt appears BEFORE the next step, and the student must type a response (minimum 5 characters) before the button to reveal the next step becomes active. This prevents clicking through without thinking.

Layout: The problem is displayed at the TOP of the screen and remains visible throughout. Steps appear below the problem in sequence — each new step appears below the previous one, so students can see the accumulating solution. Self-explanation prompts appear as inline text boxes immediately below each step. Faded steps appear as input fields with the step label ("Step 4: Substitute to find y") but the content blank.

Feedback timing: On faded steps, feedback appears immediately after the student submits their answer. Correct: green tick with brief confirmation. Incorrect: amber highlight with a specific hint (not just "wrong" but targeting the likely error). Student can retry twice before the correct answer is revealed with explanation.

Navigation: Students can scroll up to review previous steps but CANNOT skip ahead. They must complete the self-explanation prompt at each step before proceeding. They CAN go back and revise a self-explanation answer. Once the example is complete, a "Summary" view shows all steps together.

Adaptation Notes

If the platform tracks student responses:

• If a student answers all self-explanations correctly and completes faded steps on first attempt: Skip E2 (second full example) and proceed directly to E3. The student doesn't need additional full examples.
• If a student struggles with the faded steps (2+ incorrect attempts): Insert an additional faded example between E4 and E5, fading only one more step. Don't jump to full independence too quickly.
• If a student's self-explanations reveal a specific misconception: Branch to a targeted mini-lesson addressing that misconception before continuing the sequence.

---

Known Limitations

1. Worked examples are most effective for NOVICE learners. As students become proficient, worked examples become redundant and can actually hinder performance (the "expertise reversal effect" — Kalyuga et al., 2003). This sequence is designed for students' FIRST encounter with simultaneous equations. For students who already have some proficiency, start later in the fading sequence or skip to independent practice.

2. Digital self-explanation prompts can become perfunctory. Students may learn to type the minimum response to unlock the next step without genuinely engaging. The design mitigates this (minimum character count, the prompt appears before the answer) but cannot fully prevent gaming behaviour. In classroom use, the teacher should periodically review students' self-explanation responses to check for genuine engagement.

3. The fading schedule above is a fixed sequence. Truly adaptive fading would adjust the pace based on each student's performance — fading faster for students who demonstrate mastery and slower for those who struggle. The fixed schedule is a practical default for platforms without adaptive capability. Platforms with tracking should use the adaptation notes to personalise the pace.

Erroneous Example Designer

What This Skill Does

Designs worked examples that contain deliberate, realistic errors for students to identify, explain, and correct — a technique that produces learning effects comparable to or exceeding correct worked examples, with the additional benefit of developing error-detection skills. The critical insight from McLaren et al. (2012, 2015) is that errors must be REALISTIC and COMMON — the kinds of mistakes students actually make, not contrived errors that no one would make. A well-designed erroneous example activates self-explanation (Chi et al., 1989): students must reason about WHY the step is wrong, which forces deeper processing than simply following a correct procedure. The output includes the erroneous examples with realistic errors at specific steps, an error analysis scaffold (prompts that guide students to find and correct the error), the learning mechanism explanation, and the corrected version. AI is specifically valuable here because designing effective erroneous examples requires deep knowledge of the common error patterns for specific problem types — which errors are realistic, which are productively confusing, and which would create harmful misconceptions.

Evidence Foundation

McLaren, Adams & Mayer (2012) found that students who studied erroneous examples showed significantly better retention and transfer than students who studied correct examples — but this effect was DELAYED (appearing on a one-week post-test, not an immediate test). This suggests that erroneous examples produce deeper, more durable learning than correct examples, possibly because the error-detection process forces more elaborate processing. McLaren et al. (2015) replicated and extended this finding, showing that the combination of erroneous examples WITH self-explanation prompts produced the strongest effects. Tsovaltzi et al. (2010) found that erroneous examples were particularly effective when students were prompted to explain WHY the error was wrong, not just to identify it. Große & Renkl (2007) found that erroneous examples improved learning when students had sufficient prior knowledge to detect the error — but could confuse students who lacked the prerequisite knowledge (they might learn the error as correct procedure). This establishes a critical design constraint: erroneous examples work AFTER students have seen correct examples, not as first exposure. Siegler (2002) showed that children benefit from explaining both correct and incorrect strategies — the contrast between "this works and this doesn't" deepens understanding more than studying either alone.

Input Schema

The teacher must provide:

• Problem domain: The type of problem. e.g. "Adding fractions with different denominators" / "Calculating percentage increase" / "Using apostrophes for possession vs. contraction" / "Balancing chemical equations"
• Target errors: The specific common errors. e.g. "Adding numerators and denominators separately: ½ + ⅓ = 2/5" / "Calculating percentage OF the increase rather than percentage increase: confusing 'what is 20% of 80?' with 'what is the percentage increase from 80 to 96?'" / "Using an apostrophe for plurals: apple's instead of apples" / "Changing coefficients into subscripts when balancing"

Optional (injected by context engine if available):

• Student level: Year group and proficiency
• Subject area: Curriculum subject
• Correct examples available: Whether students have seen correct versions
• Number of examples: How many to design
• Delivery context: Digital, paper, or class discussion

Prompt

You are an expert in erroneous example design for learning, with deep knowledge of McLaren et al.'s (2012, 2015) research on delayed learning effects from erroneous examples, Tsovaltzi et al.'s (2010) work on error-based learning, Große & Renkl's (2007) research on finding and fixing errors in worked examples, and Siegler's (2002) microgenetic studies of self-explanation with correct and incorrect strategies. You understand that erroneous examples are not "trick questions" — they are carefully designed learning tools where realistic, common errors are embedded at specific steps, and students learn by DETECTING, EXPLAINING, and CORRECTING the error.

CRITICAL PRINCIPLES:
- **Errors must be REALISTIC and COMMON.** The error should be one that students actually make — a genuine misconception or procedural slip, not an absurd mistake. "3 + 4 = 12" is not a realistic error. "½ + ⅓ = 2/5" IS a realistic error (adding numerators and denominators separately). Realistic errors activate recognition: "I've made this mistake" or "I can see why someone would think that."
- **One error per example.** An example with multiple errors is confusing, not instructive. Embed ONE error at ONE specific step, with all other steps correct. This isolates the learning target and makes detection feasible.
- **Students must already have seen correct examples.** Große & Renkl (2007) showed that erroneous examples confuse students who haven't seen correct versions first. Use erroneous examples AFTER correct worked examples, not instead of them. The sequence is: correct examples → erroneous examples → independent practice.
- **The error analysis scaffold is essential.** Simply showing an erroneous example is insufficient. Students need prompts: "Find the error," "Explain why it's wrong," "Correct it," "Explain why your correction is right." This scaffold forces the self-explanation that produces the learning effect.
- **Erroneous examples develop error-detection skills.** Beyond learning the specific content, students who practise with erroneous examples become better at monitoring their OWN work for errors. This metacognitive benefit is separate from and additional to the content learning.

Your task is to design erroneous examples for:

**Problem domain:** {{problem_domain}}
**Target errors:** {{target_errors}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Student level:** {{student_level}} — if not provided, design for a general secondary school context.
**Subject area:** {{subject_area}} — if not provided, infer from the problem domain.
**Correct examples available:** {{correct_examples_available}} — if not provided, include a note that students should see correct examples first.
**Number of examples:** {{number_of_examples}} — if not provided, design 3 erroneous examples targeting different common errors.
**Delivery context:** {{delivery_context}} — if not provided, design for paper-based use that could be adapted for digital delivery.

Return your output in this exact format:

## Erroneous Examples: [Problem Domain]

**Problem domain:** [The type of problem]
**Target errors:** [The common errors being addressed]
**Prerequisite:** [What students must already know — correct examples must come first]

### Erroneous Example [N]

**The problem:** [The problem being solved]
**The erroneous solution:**
[Step-by-step solution with ONE deliberate error at a specific step — all other steps correct]

**The error:** [Which step is wrong and what the error is — FOR TEACHER REFERENCE ONLY, not shown to students]
**Why this error is realistic:** [Why students actually make this mistake — the underlying misconception or procedural confusion]

**Error Analysis Scaffold (for students):**
1. "Read through the solution carefully. Is every step correct?"
2. "Find the step that contains an error. Circle it."
3. "Explain WHY this step is wrong. What mistake was made?"
4. "Write the correct version of this step."
5. "Complete the problem correctly from this point."

### Correct Version (Teacher Reference)

[The fully correct solution for comparison]

### Learning Mechanism

[Why these erroneous examples produce learning — the cognitive process of error detection, explanation, and correction]

### Sequencing Guidance

[When to use these erroneous examples in a lesson — after correct examples, before independent practice]

**Self-check before returning output:** Verify that (a) each error is realistic and common, (b) there is only ONE error per example, (c) the error analysis scaffold requires explanation not just identification, (d) correct versions are provided for teacher reference, and (e) the examples are sequenced after correct worked examples.

Example Output

Scenario: Problem domain: "Adding fractions with different denominators — Year 6 Mathematics" / Target errors: "Error 1: Adding numerators and denominators separately (½ + ⅓ = 2/5). Error 2: Finding a common denominator but forgetting to adjust the numerators (½ + ⅓ = 1/6 + 1/6 = 2/6). Error 3: Correctly finding equivalent fractions but making an arithmetic error in the addition (3/6 + 2/6 = 6/6 instead of 5/6)." / Correct examples available: "Yes — students have completed correct worked examples of adding fractions with common denominators and have seen the procedure for finding common denominators"

---

Erroneous Examples: Adding Fractions With Different Denominators

Problem domain: Adding fractions with different denominators Target errors: Adding numerators/denominators separately, forgetting to adjust numerators, arithmetic errors in final addition Prerequisite: Students must have already studied correct worked examples of finding common denominators and adding equivalent fractions. Erroneous examples are used AFTER correct examples to deepen understanding.

Erroneous Example 1: "Sam's Solution"

The problem: Calculate ½ + ⅓

Sam's solution:

• Step 1: I need to add ½ and ⅓
• Step 2: I add the numerators: 1 + 1 = 2
• Step 3: I add the denominators: 2 + 3 = 5
• Step 4: My answer is 2/5 ✓

The error (teacher reference): Step 2 and 3. Sam adds numerators and denominators separately, treating fraction addition like whole number addition. This is the most common fraction error and reflects a fundamental misconception about what fractions ARE — the denominator names the size of the parts, and you cannot add parts of different sizes by adding the size-names.

Why this error is realistic: Students who have just learned to add whole numbers extend the same logic to fractions. It "looks right" — you're adding the tops and adding the bottoms, just like you'd add two whole numbers. The error is so common that research consistently finds it in 30–50% of students learning fraction addition.

Error Analysis Scaffold (for students):

1. "Read Sam's solution. Does each step make sense?"
2. "Sam says ½ + ⅓ = 2/5. Think about this: ½ is the same as 0.5. Is 2/5 bigger or smaller than ½?" (2/5 = 0.4, which is SMALLER than ½ alone. This is impossible — adding a positive fraction to ½ must give a result bigger than ½.)
3. "Which step went wrong? Circle it and explain WHY it's wrong."
4. "What should Sam have done instead? Solve the problem correctly."
5. "Write Sam a note explaining the mistake in a way that would help them understand."

Erroneous Example 2: "Priya's Solution"

The problem: Calculate ½ + ⅓

Priya's solution:

• Step 1: I need a common denominator. 2 and 3 both go into 6. The common denominator is 6.
• Step 2: ½ = 1/6 (I changed the denominator to 6)
• Step 3: ⅓ = 1/6 (I changed the denominator to 6)
• Step 4: 1/6 + 1/6 = 2/6
• Step 5: My answer is 2/6 ✓

The error (teacher reference): Steps 2 and 3. Priya correctly finds the common denominator (6) but changes ONLY the denominator without adjusting the numerator. She writes ½ = 1/6, but ½ = 3/6 (multiply both top and bottom by 3). Similarly, ⅓ = 2/6 (multiply both top and bottom by 2). Priya understands that she needs a common denominator but has not grasped that the fraction must remain EQUIVALENT — both parts must change by the same factor.

Why this error is realistic: Students who have been taught "find a common denominator" sometimes treat it as changing the bottom number without understanding that the fraction must stay equivalent. They follow the PROCEDURE (change the denominator) without understanding the PRINCIPLE (maintain equivalence). This is a procedural-without-conceptual error.

Error Analysis Scaffold (for students):

1. "Priya found the correct common denominator (6). Good start. But look at Step 2."
2. "Priya says ½ = 1/6. Use a diagram or fraction strips: is ½ the same as 1/6?" (No — ½ is much bigger than 1/6.)
3. "What went wrong? When Priya changed ½ to sixths, what should the numerator be? How do you know?"
4. "Correct Priya's work from Step 2 onwards."
5. "What RULE should Priya remember? (Hint: what you do to the bottom, you must also do to the ___)"

Erroneous Example 3: "Marcus's Solution"

The problem: Calculate ½ + ⅓

Marcus's solution:

• Step 1: I need a common denominator. 2 and 3 both go into 6.
• Step 2: ½ = 3/6 (multiply top and bottom by 3) ✓
• Step 3: ⅓ = 2/6 (multiply top and bottom by 2) ✓
• Step 4: 3/6 + 2/6 = 6/6
• Step 5: 6/6 = 1 whole
• Step 6: My answer is 1 ✓

The error (teacher reference): Step 4. Marcus does everything correctly until the final addition: 3/6 + 2/6 = 5/6, not 6/6. This is a simple arithmetic error (3 + 2 = 5, not 6), but it leads to a plausible-looking final answer (1 whole) that might not trigger the student's error-detection because it "looks reasonable."

Why this error is realistic: Once students have successfully navigated the conceptually hard parts (finding common denominator, adjusting numerators), they can make careless arithmetic errors at the final step. Because the difficult conceptual work is done, students relax their monitoring and miss the simple mistake. Additionally, 6/6 = 1 looks like a "neat" answer, which confirms the error rather than flagging it.

Error Analysis Scaffold (for students):

1. "Marcus found the common denominator correctly and converted both fractions correctly. Well done so far."
2. "Look carefully at Step 4. Check the addition: 3 + 2 = ?"
3. "What should the correct answer be? Write the corrected step."
4. "Marcus's error was small — an arithmetic slip. But it changed the answer completely. What does this tell you about checking your work?"

Correct Version (Teacher Reference)

½ + ⅓ = 3/6 + 2/6 (common denominator 6; ½ × 3/3 = 3/6; ⅓ × 2/2 = 2/6) = 5/6

Learning Mechanism

These three erroneous examples target three DIFFERENT types of error at three different points in the procedure:

1. Example 1 (Sam): A CONCEPTUAL error — misunderstanding what fraction addition means. Addresses the most fundamental misconception.
2. Example 2 (Priya): A PROCEDURAL-CONCEPTUAL gap — following the procedure without understanding why both numerator and denominator must change. Addresses the disconnect between procedure and understanding.
3. Example 3 (Marcus): An ARITHMETIC error — everything conceptually correct but a careless calculation mistake. Addresses monitoring and checking.

By presenting all three, students develop a taxonomy of error types: "When I add fractions, I need to check for conceptual errors (am I adding correctly?), procedural errors (did I adjust the numerators?), AND arithmetic errors (did I add the numbers right?)."

Sequencing Guidance

Lesson position: Use these erroneous examples AFTER students have:

1. Learned the correct procedure through correct worked examples
2. Practised 2–3 problems with guided support

Introduce as: "Here are three students' solutions. Each one has made a mistake. Your job is to find the mistake, explain what went wrong, and fix it."

Follow with: Independent practice where students solve problems themselves — now with the error-detection skills to monitor their own work.

---

Known Limitations

1. Erroneous examples can create misconceptions if used before correct examples (Große & Renkl, 2007). Students who encounter errors before they have a secure model of the correct procedure may inadvertently learn the error as correct. The sequencing is critical: correct examples FIRST, then erroneous examples to deepen understanding and develop error-detection skills.

2. The learning effect is often DELAYED (McLaren et al., 2012). Students who study erroneous examples may not outperform students who study correct examples on an immediate test — but they show superior retention and transfer on delayed tests (one week later). Teachers should be aware that the benefit may not be immediately visible, and should not conclude the approach has failed based on a same-day assessment.

3. The quality of the error-analysis scaffold determines the learning effect. Simply showing students an error and saying "find the mistake" produces much weaker effects than providing structured prompts that require explanation and correction (Tsovaltzi et al., 2010). The scaffold above is designed to force self-explanation — but if students skip the explanation step and just identify the error without reasoning about it, the learning benefit is significantly reduced.

Formative Assessment Loop Designer for AI Systems

What This Skill Does

Designs a complete formative assessment loop for an AI-enabled learning environment — the continuous cycle of eliciting evidence of student understanding, interpreting that evidence, and using it to adjust instruction in real time. Black & Wiliam's (1998) seminal meta-analysis demonstrated that formative assessment is one of the most powerful interventions in education (effect sizes of 0.40-0.70), but ONLY when the assessment data actually changes what happens next. An assessment that doesn't lead to an instructional adjustment is just a test. This skill designs the entire loop: what to assess (not just answers but THINKING), how to assess it (questions, tasks, and probes that reveal understanding), how to interpret the results (distinguishing genuine understanding from surface performance), and what to do with the results (specific instructional responses to specific assessment patterns). VanLehn (2006) distinguished between "inner loop" assessment (at each problem step, in real time) and "outer loop" assessment (at the task level, between problems). AI systems are uniquely capable of inner-loop assessment — monitoring student reasoning step by step and adjusting in real time — which is where the largest learning gains occur.

Evidence Foundation

Black & Wiliam (1998) conducted the most influential review of formative assessment research, analysing over 250 studies and finding effect sizes ranging from 0.40 to 0.70 — larger than most educational interventions. They defined formative assessment as "all those activities undertaken by teachers, and/or by their students, which provide information to be used as feedback to modify the teaching and learning activities in which they are engaged." The critical insight: the ASSESSMENT is not the valuable part — the MODIFICATION is. Data collection without instructional adjustment is summative assessment with a different label. Black & Wiliam (2009) developed a more refined theoretical framework, identifying five key strategies of formative assessment: (1) clarifying and sharing learning intentions and criteria for success, (2) engineering effective classroom discussions and tasks that elicit evidence of learning, (3) providing feedback that moves learners forward, (4) activating students as instructional resources for one another, and (5) activating students as owners of their own learning. All five strategies involve a cycle: elicit → interpret → act. Wiliam (2011) translated this framework into practical classroom strategies, emphasising that formative assessment must be EMBEDDED in instruction — not an add-on activity but a continuous process of checking, adjusting, and responding. He argued that the biggest barrier to effective formative assessment is not data collection but DATA USE: teachers often collect data but don't change their teaching in response to it. VanLehn (2006) analysed the behaviour of tutoring systems and identified two levels of assessment loop. The "outer loop" operates between problems: after a student completes a problem, the system decides what to do next (another similar problem, a harder problem, a review, or a new topic). The "inner loop" operates within problems: at each step, the system assesses the student's response and provides feedback, hints, or scaffolding. VanLehn found that the inner loop was the more important determinant of ITS effectiveness — systems that assessed and responded at the step level dramatically outperformed systems that only assessed at the problem level. Shute & Zapata-Rivera (2012) reviewed adaptive educational systems and found that the most effective systems combined continuous assessment with immediate instructional adaptation — creating a "tight" formative assessment loop where the time between assessment and response was minimised.

Input Schema

The teacher must provide:

• Learning objective: What students are trying to learn. e.g. "Understanding the difference between area and perimeter — when to use each, how to calculate each, and why they are independent (a shape can have a large area but small perimeter, and vice versa)" / "Writing effective topic sentences that make a clear claim, not just state a fact or announce a topic" / "Understanding supply and demand — how price changes affect quantity supplied and demanded, and how equilibrium price is determined"
• Current assessment approach: How understanding is currently checked. e.g. "I check at the end of the lesson with an exit ticket — 3 questions. If most students get them right, I move on. If not, I reteach next lesson" / "I mark essays at the end of the week and give written feedback" / "The AI system gives a quiz after every 10 problems, but doesn't change what happens next based on the results"

Optional (injected by context engine if available):

• Student level: Year group and proficiency
• Subject area: The curriculum subject
• AI system capabilities: What the technology can do
• Class size: Number of students
• Assessment frequency: How often to collect data

Prompt

You are an expert in formative assessment design for AI-enabled learning environments, with deep knowledge of Black & Wiliam's (1998, 2009) formative assessment framework, Wiliam's (2011) practical implementation strategies, VanLehn's (2006) inner-loop/outer-loop distinction, and Shute & Zapata-Rivera's (2012) adaptive assessment systems. You understand that formative assessment is not a type of test — it is a PROCESS of continuously eliciting evidence of understanding and using that evidence to adjust instruction. You also understand VanLehn's critical finding: assessment and feedback at the STEP level (inner loop) is dramatically more effective than assessment at the TASK level (outer loop).

CRITICAL PRINCIPLES:
- **Assessment must change instruction.** If the assessment data doesn't lead to a different instructional response, it's not formative — it's just a test. For every assessment point, specify WHAT CHANGES based on the result. "If the student gets it right, move on" is insufficient. "If the student gets it right, increase difficulty by X; if wrong in way A, respond with action A; if wrong in way B, respond with action B" — that's formative.
- **Assess THINKING, not just answers.** A correct answer might hide a misconception (right answer, wrong reasoning). An incorrect answer might contain valuable partial understanding. The assessment must probe the reasoning BEHIND the answer. In an AI system: require students to show working, explain their thinking, or select from options that reveal specific reasoning patterns.
- **Inner-loop assessment is more powerful than outer-loop.** VanLehn (2006): assessing at each problem step (and responding immediately) produces better learning than assessing only at the end of a problem. Design the loop to operate at the step level where possible.
- **Use multiple elicitation methods.** Don't rely solely on correct/incorrect. Use: diagnostic questions (MCQs where each wrong answer maps to a specific misconception), explanation prompts ("Why did you choose that?"), confidence ratings ("How sure are you?"), and process observations (how long did they take? did they use a hint?).
- **The assessment loop must be TIGHT.** The shorter the delay between assessment and instructional response, the more effective the formative process. An AI system can respond in seconds. Exploit this advantage — don't collect data now and act on it next week.

Your task is to design a formative assessment loop for:

**Learning objective:** {{learning_objective}}
**Current assessment approach:** {{current_assessment_approach}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Student level:** {{student_level}} — if not provided, design for a general secondary school context.
**Subject area:** {{subject_area}} — if not provided, infer from the objective.
**AI system capabilities:** {{ai_system_capabilities}} — if not provided, design for an AI system that can present problems, monitor responses, provide feedback, and adapt problem selection in real time.
**Class size:** {{class_size}} — if not provided, assume a class of 30 students working individually on AI-enabled devices.
**Assessment frequency:** {{assessment_frequency}} — if not provided, design for continuous inner-loop assessment with outer-loop checks every 5-10 problems.

Return your output in this exact format:

## Formative Assessment Loop: [Learning Objective]

**Objective:** [What students are learning]
**Current approach:** [How assessment works now]
**Redesigned approach:** [How the formative loop works — one-sentence summary]

### Loop Architecture

[The complete assessment loop structure — inner loop (step-level) and outer loop (task-level)]

**Inner Loop (within each problem):**
[What is assessed at each step, how, and what the response is]

**Outer Loop (between problems):**
[What is assessed after each problem/set of problems, and how it determines what comes next]

### Elicitation Strategies

[The specific methods for surfacing student understanding — not just "quiz them" but the SPECIFIC question designs, process observations, and explanation prompts]

**Strategy [N]: [Name]**
- **What it assesses:** [What aspect of understanding]
- **How it works:** [The specific mechanism]
- **Example:** [A concrete example for this learning objective]
- **What to look for:** [What different responses reveal about understanding]

### Interpretation Framework

[How to interpret student responses — the decision rules that connect assessment evidence to instructional actions]

| Evidence Pattern | What It Probably Means | Confidence | Instructional Response |
|---|---|---|---|
| [Pattern] | [Interpretation] | [High/Moderate/Low] | [What the system does next] |

### Response Actions

[The specific instructional adjustments triggered by different assessment results — not just "reteach" but exactly how to reteach]

### Teacher Dashboard

[What the teacher needs to see — the key metrics and alerts that inform teacher-level decisions, separate from the AI's real-time responses]

### Loop Validation

[How to check that the assessment loop is actually improving learning — the meta-assessment of the assessment]

**Self-check before returning output:** Verify that (a) every assessment point triggers an instructional response, (b) thinking is assessed not just answers, (c) inner-loop assessment is included, (d) multiple elicitation methods are used, and (e) the loop is tight (minimal delay between assessment and response).

Example Output

Scenario: Learning objective: "Understanding the difference between area and perimeter — when to use each, how to calculate each, and why they are independent (a shape can have a large area but small perimeter, and vice versa)" / Current assessment approach: "I teach area and perimeter in separate lessons, then give a worksheet with 20 calculation questions (10 area, 10 perimeter). I mark them and reteach if more than a third of the class got fewer than 7/10 on either section. The main problem is that students can calculate both correctly on the worksheet but CONFUSE them on tests — they use the perimeter formula when asked for area, or vice versa. Calculation isn't the issue; knowing WHICH calculation to use is." / Student level: "Year 6" / Subject area: "Mathematics — measurement" / AI system capabilities: "AI tutoring system that can present problems, accept free-text and multiple-choice responses, time student responses, deliver feedback, and adapt problem selection. Can also show images/diagrams of shapes." / Assessment frequency: "Continuous"

---

Formative Assessment Loop: Area vs. Perimeter

Objective: Understanding when to use area vs. perimeter, how to calculate each, and why they are independent Current approach: Separate teaching → mixed calculation worksheet → mark and reteach Redesigned approach: Continuous assessment loop that probes WHICH calculation to use (not just whether the calculation is correct), using diagnostic questions that separate conceptual understanding from procedural fluency

Loop Architecture

Inner Loop (within each problem):

Step 1 — Identification check: Before the student calculates anything, the system asks: "Is this an area question or a perimeter question? How do you know?" This is the CRITICAL assessment point. The most common error (using the wrong formula) occurs at THIS step, not during calculation. The system assesses the student's conceptual identification before allowing them to proceed.

Step 2 — Method check: After identification, the system asks: "What will you calculate? Describe your method in one sentence." This reveals whether the student knows the procedure for the identified concept. A student who correctly identifies "area" but describes "I'll add up all the sides" has a procedural error despite correct identification.

Step 3 — Calculation check: The student performs the calculation. The system checks the numerical answer. If correct, proceed. If incorrect, determine whether the error is arithmetic (calculation mistake with correct method) or conceptual (wrong method applied).

Step 4 — Units check: The system checks whether the student has used correct units (cm² for area, cm for perimeter). Incorrect units may reveal conceptual confusion even when the numerical answer is correct.

Outer Loop (every 5 problems):

After every 5 problems, the system reviews the pattern of inner-loop responses:

• If identification accuracy ≥ 80%: Advance difficulty (more complex shapes, word problems, comparison questions)
• If identification accuracy < 60%: Switch to concept-building tasks (compare shapes with same perimeter but different areas)
• If identification accurate but calculation often wrong: Provide calculation practice with the correctly identified concept
• If mixed pattern: Present a diagnostic question designed to pinpoint the specific confusion

Elicitation Strategies

Strategy 1: Identification-First Questions

• What it assesses: Whether the student knows WHICH measurement is being asked for
• How it works: Present a word problem and ask "Is this asking for area or perimeter?" BEFORE asking for the calculation. This separates conceptual identification from procedural execution.
• Example: "A farmer wants to build a fence around a rectangular field that is 20m long and 15m wide. Is this an area question or a perimeter question?" Then, separately: "Calculate the answer."
• What to look for: Students who correctly identify "perimeter" (it's about fencing = going around the outside) vs. students who say "area" (they may be triggered by the word "field" or the two dimensions)

Strategy 2: Diagnostic Multiple Choice

• What it assesses: The specific misconception behind a wrong answer
• How it works: Present a problem where each wrong answer maps to a specific error. The student's choice reveals their reasoning, not just their accuracy.
• Example: "A rectangle is 8cm long and 5cm wide. What is its area?" Options: (a) 40 cm² [correct], (b) 26 cm [calculated perimeter instead], (c) 26 cm² [calculated perimeter but used area units], (d) 13 cm [calculated half-perimeter]. Each wrong answer diagnoses a different problem.
• What to look for: (b) = calculated the wrong measure. (c) = calculated the wrong measure AND doesn't understand what the units mean. (d) = partial perimeter understanding.

Strategy 3: Concept Comparison Tasks

• What it assesses: Whether the student understands that area and perimeter are INDEPENDENT
• How it works: Show two shapes and ask "Which has the bigger area? Which has the bigger perimeter?" where the answers are DIFFERENT (one has bigger area, the other has bigger perimeter). This tests the deepest conceptual understanding.
• Example: Shape A: 10cm × 2cm rectangle (area 20, perimeter 24). Shape B: 5cm × 5cm square (area 25, perimeter 20). "Which has the bigger area? Which has the bigger perimeter?" Correct: B has bigger area, A has bigger perimeter.
• What to look for: Students who say the same shape for both questions probably believe area and perimeter are correlated ("bigger shape = bigger everything"). This is the most important misconception to address.

Strategy 4: Confidence-Calibrated Responses

• What it assesses: Metacognitive accuracy — whether the student knows what they know
• How it works: After answering, student rates confidence (sure / think so / guessing). The system tracks calibration: high confidence + wrong answer = misconception (they believe something incorrect). Low confidence + right answer = fragile knowledge (they got lucky or are uncertain).
• Example: Student confidently says "perimeter" for a fencing question → correct and confident = solid understanding. Student uncertainly says "perimeter" → correct but uncertain = developing understanding, needs more practice.
• What to look for: Consistent high-confidence errors signal firmly held misconceptions that need direct confrontation, not just more practice.

Interpretation Framework

Evidence Pattern	What It Probably Means	Confidence	Instructional Response
Correct identification, correct calculation, correct units	Solid understanding of this problem type	High	Increase difficulty — more complex shapes, multi-step problems, or comparison tasks
Correct identification, wrong calculation	Procedural gap, not conceptual confusion	High	Provide calculation scaffolding — the student knows WHAT to do but makes arithmetic errors
Wrong identification (area↔perimeter confusion)	Core conceptual confusion — the student doesn't reliably distinguish the two concepts	High	STOP calculation practice. Switch to concept-building: comparisons, definitions, visual demonstrations, real-world contexts
Correct identification on simple problems, wrong on word problems	Fragile conceptual knowledge — can identify in explicit contexts but not in applied contexts	Moderate	More word problems with explicit identification prompts ("Is this about covering a surface or going around the edge?")
Correct answer but wrong units (e.g., 40 cm instead of 40 cm²)	May understand the concept but not the mathematical representation of dimensions	Moderate	Direct teaching on why area uses squared units (it measures the number of unit squares)
Same answer for "bigger area" and "bigger perimeter" comparison	Believes area and perimeter are correlated	High	Key intervention: multiple examples showing shapes where area increases but perimeter decreases (and vice versa)
High confidence + consistent errors	Firmly held misconception	High	Cognitive conflict: present a case that contradicts their belief and ask them to explain the contradiction
Low confidence + correct answers	Developing but fragile understanding	Moderate	More practice with immediate positive feedback to build confidence alongside competence

Response Actions

Action 1: Concept-Building Mode (triggered by identification errors) The AI switches from calculation practice to concept-building activities:

• Show a rectangle and ask "What does area mean for this shape? What does perimeter mean? Show me by pointing/highlighting." (Area = inside surface, perimeter = outside edge)
• Present real-world scenarios: "You want to paint a wall — area or perimeter? You want to put a frame around a picture — area or perimeter? You want to buy carpet for a room — area or perimeter?"
• Use the "same perimeter, different area" demonstration: show three shapes all with perimeter 24cm but very different areas (1×11, 4×8, 6×6). Ask: "How can the perimeter be the same but the areas be different?"

Action 2: Procedural Consolidation Mode (triggered by correct identification but calculation errors) The AI provides scaffolded calculation practice with the CORRECTLY identified concept:

• Area: Show grid overlays so students can count squares before calculating. Fade the grid.
• Perimeter: Show arrows around the outside of the shape before students add. Fade the arrows.
• Both: Provide immediate feedback on the calculation with corrective steps shown.

Action 3: Integration Mode (triggered by consistent success) The AI introduces multi-step and comparison problems:

• "Find BOTH the area and perimeter of this shape. Which is larger? Does that surprise you?"
• "Design a rectangle with an area of 24cm². What perimeter does it have? Can you design a different rectangle with the same area but a different perimeter?"
• Word problems that require the student to determine which measurement is needed from context.

Teacher Dashboard

The teacher needs to see:

Class-level overview:

• % of students in each mode (concept-building / procedural / integration)
• The most common error pattern across the class (if 60% are confusing area and perimeter, this needs a whole-class intervention)
• Students who are stuck (no improvement after 10+ problems in concept-building mode — may need teacher intervention)

Individual alerts:

• "Student X has high confidence but 40% identification accuracy — firmly held misconception, may benefit from teacher conversation"
• "Student Y has been in procedural mode for 15 minutes — calculation accuracy not improving, check for underlying number sense gap"
• "Student Z completed all integration tasks — ready for extension (compound shapes, circles, surface area)"

Key metric: Identification accuracy vs. calculation accuracy. If the class identification accuracy is below 70% but calculation accuracy is above 80%, the current assessment approach (calculation worksheets) is masking the real problem. The teacher should prioritise conceptual work over more calculation practice.

Loop Validation

How to check that the assessment loop is working:

1. Pre-post comparison on identification tasks. Before the AI loop begins, give a 5-question identification-only quiz (no calculation). After 2 sessions with the AI loop, repeat the quiz. If identification accuracy hasn't improved, the loop's concept-building responses need redesigning.

2. Transfer test. After the AI sequence, give a novel problem type that the AI didn't use (e.g., estimating whether a piece of wrapping paper is big enough for a present — requires understanding of area in an unfamiliar context). If students can transfer, the loop is building understanding, not just training on familiar problem formats.

3. Delayed test. Two weeks after the AI sequence, give the same identification quiz. If scores have dropped significantly, the loop produced SHORT-TERM performance but not LONG-TERM learning. Consider adding spaced retrieval to the loop (revisiting area/perimeter identification occasionally after the student has moved on).

---

Known Limitations

1. Formative assessment loops require continuous data flow. The design above assumes the AI system can assess student responses in real time and adapt immediately. Systems with batch processing (collect data, analyse overnight, adjust tomorrow) cannot implement the inner loop. The outer loop is still possible and still valuable, but the learning gains from inner-loop assessment (VanLehn, 2006) require real-time processing.

2. The interpretation framework is probabilistic. A student who chooses "area" for a perimeter question PROBABLY has a conceptual confusion — but they might have misread the question, clicked the wrong option, or not understood the word "perimeter." The system should never make a definitive diagnosis from a single response. The interpretation framework above uses PATTERNS of responses (multiple problems) to build confidence in the diagnosis.

3. Black & Wiliam's (1998) meta-analysis included a wide range of formative assessment practices. The effect sizes (0.40-0.70) apply to formative assessment broadly, not specifically to AI-implemented formative assessment. The principles are sound, but the specific effect size of an AI formative loop in THIS context has not been empirically measured. The design is evidence-informed, not evidence-proven.

4. Teacher dashboard data can be overwhelming. The dashboard above provides detailed information about individual students and class-level patterns. In a class of 30, this is manageable. In a year group of 120 students, the data volume may be overwhelming. Dashboard design for larger scales requires more aggressive filtering and summarisation.

5. The assessment loop may inadvertently narrow the curriculum. If the AI loop focuses exclusively on area vs. perimeter identification and calculation, students may develop competence in this specific skill but miss the broader mathematical understanding (measurement as a concept, connections to other topics). Wiliam (2011) warns that formative assessment should serve learning, not define it.

Individual Spacing Algorithm Explainer

What This Skill Does

Generates a personalised spaced repetition schedule for a specific set of knowledge items based on student performance data, and explains the algorithm logic in teacher-friendly terms. This skill does NOT execute the spacing algorithm — execution requires a system like Anki, Quizlet, SuperMemo, or Kaku's own retention engine. What this skill does is DESIGN and EXPLAIN: given what students know and what they're forgetting, it produces a concrete review schedule with clear rationale for why each interval was chosen. The core insight from spacing research is that the optimal review interval depends on the desired retention period and the item's current memory strength. Ebbinghaus (1885) established that forgetting follows a predictable curve — steep initially, then gradually flattening. Cepeda et al. (2006) demonstrated in a comprehensive meta-analysis that spacing review across time produces substantially better long-term retention than massing the same amount of practice into a single session. The practical challenge is that the optimal interval is different for every student and every item — which is where personalisation becomes essential. AI is specifically valuable here because calculating optimal intervals across dozens of items for individual students is computationally trivial for a machine but practically impossible for a teacher doing it manually.

Evidence Foundation

Ebbinghaus (1885/1913) conducted the foundational experiments on forgetting, demonstrating that memory for newly learned material decays exponentially but that each subsequent review strengthens the memory trace and slows the rate of forgetting. This "forgetting curve" is the theoretical foundation for all spacing algorithms. Cepeda et al. (2006) conducted a meta-analysis of 254 studies on distributed practice, finding a robust spacing effect across a wide range of materials and age groups. They found that the optimal inter-study interval (ISI) depends on the retention interval (RI) — the time until the knowledge is needed. A rough guideline from their analysis: the optimal ISI is approximately 10-20% of the RI. So if you need knowledge for an exam in 30 days, optimal spacing is roughly every 3-6 days. Lindsey et al. (2014) moved spacing research into real classrooms, testing a personalised spacing algorithm with middle school students learning social studies. Their algorithm adjusted review intervals based on individual student performance, and they found significant improvements in long-term retention compared to massed review schedules chosen by teachers. Critically, the personalised algorithm outperformed a one-size-fits-all spacing schedule, demonstrating the value of individual calibration. Settles & Meeder (2016) developed the "half-life regression" model used in Duolingo, which predicts the probability that a student will recall a specific item at a specific time based on their individual history with that item. The model combines three variables: the number of times the item has been seen, the time since last review, and the student's accuracy on similar items. This represents the state of the art in practical, large-scale spacing algorithms. Pashler et al. (2007) translated spacing research into practical recommendations in an IES practice guide, noting that teachers rarely use spacing despite its robust evidence base — partly because it requires advance planning and partly because it feels counterintuitive (students feel more confident after massed practice, even though spaced practice produces better retention).

Input Schema

The teacher must provide:

• Content to space: What students need to retain. e.g. "Year 8 biology: cell organelles and their functions — mitochondria, nucleus, cell membrane, ribosomes, chloroplasts, vacuole, cell wall" / "Key dates and causes of WWI" / "French irregular verb conjugations — être, avoir, aller, faire" / "Times tables 6-12"
• Performance data: What is known about current retention. e.g. "End-of-topic quiz last Friday: class average 72%. Most students got mitochondria and nucleus correct but confused ribosomes with rough ER. Three students scored below 50%" / "Informal questioning suggests most students remember causes but not dates" / "Duolingo-style accuracy: être 95%, avoir 88%, aller 71%, faire 62%"

Optional (injected by context engine if available):

• Student level: Year group and proficiency
• Subject area: The curriculum subject
• Time horizon: How long the knowledge needs to be retained
• Available review slots: How many review sessions per week are realistic
• System context: Manual, AI-embedded, or tool-based implementation

Prompt

You are an expert in the cognitive science of memory and spaced repetition, with deep knowledge of Ebbinghaus's (1885) forgetting curve, Cepeda et al.'s (2006) meta-analysis of distributed practice, Lindsey et al.'s (2014) personalised spacing research, Settles & Meeder's (2016) half-life regression model from Duolingo, and Pashler et al.'s (2007) IES practice guide on organising instruction. You understand both the theory (why spacing works — memory consolidation, retrieval practice, desirable difficulty) and the practical challenge (designing a schedule that a real teacher can actually implement).

CRITICAL PRINCIPLES:
- **The forgetting curve is exponential, but each retrieval slows it.** New material needs review soon (within 1-2 days). Each successful retrieval strengthens the memory trace and extends the optimal interval. Items retrieved 5+ times can go weeks or months between reviews.
- **The optimal interval depends on the retention horizon.** Cepeda et al.'s (2006) guideline: the optimal inter-study interval is roughly 10-20% of the time until the knowledge is needed. Short-term retention (exam next week) = short intervals. Long-term retention (GCSE in 2 years) = longer intervals with gradual expansion.
- **Not all items need the same schedule.** Items the student already knows well need less frequent review. Items they are forgetting need urgent, short-interval review. The schedule must be PERSONALISED to the performance data.
- **Prioritise by forgetting risk, not by importance.** A student who remembers mitochondria perfectly doesn't need to review mitochondria, even if it's "important." The schedule should target the items with the HIGHEST probability of being forgotten before the next review opportunity.
- **Explain the algorithm in plain English.** Teachers are not data scientists. The explanation must be clear, concrete, and jargon-free. "Review ribosomes on Tuesday because your students got it wrong last Friday and the forgetting curve suggests they'll have forgotten it by Wednesday" — not "Apply exponential decay function with λ = 0.3."

Your task is to design a personalised spacing schedule for:

**Content to space:** {{content_to_space}}
**Performance data:** {{performance_data}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Student level:** {{student_level}} — if not provided, design for a general secondary school context.
**Subject area:** {{subject_area}} — if not provided, infer from the content.
**Time horizon:** {{time_horizon}} — if not provided, assume the knowledge needs to be retained for at least one term (10-12 weeks).
**Available review slots:** {{available_review_slots}} — if not provided, assume 2-3 short review opportunities per week (5-10 minutes each).
**System context:** {{system_context}} — if not provided, design for a teacher implementing the schedule manually (low-tech, practical).

Return your output in this exact format:

## Personalised Spacing Schedule: [Content Area]

**Content:** [What needs to be retained]
**Performance snapshot:** [Summary of current retention state]
**Time horizon:** [How long the knowledge needs to last]
**Schedule basis:** [Brief explanation of why these intervals were chosen]

### Forgetting Risk Analysis

[For each item or item cluster, estimate the current forgetting risk based on the performance data. Rank from highest risk to lowest. Explain in plain English why each item is where it is.]

### Review Schedule

[A concrete, week-by-week schedule showing when each item should be reviewed. Include the format of review (quiz, flashcard, retrieval practice, application task) and the rationale for each timing decision.]

**Week 1:**
- [Day]: [Items to review] — [Why now, what format]

**Week 2:**
- [Day]: [Items to review] — [Why now, what format]

[Continue for the full time horizon or until the schedule stabilises into a maintenance pattern]

### Algorithm Explanation (Teacher-Friendly)

[Explain how the schedule was designed — in plain English. Cover: why items are reviewed at different intervals, why some items appear more often than others, how the schedule adapts if students get items right or wrong, and the underlying principle (each successful retrieval extends the interval).]

### Adaptation Rules

[What to do when students get items RIGHT (extend the interval) and when they get items WRONG (shorten the interval). Provide specific, actionable rules a teacher can follow.]

### Implementation Tips

[Practical advice for actually running this schedule in a classroom — including how to keep it manageable, how to adjust when time is short, and common mistakes to avoid.]

**Self-check before returning output:** Verify that (a) the schedule is personalised to the performance data, (b) high-forgetting-risk items are scheduled more frequently, (c) intervals expand over time for well-retained items, (d) the algorithm explanation is in plain English, and (e) the schedule is realistic for the available review time.

Example Output

Scenario: Content to space: "Year 8 biology: cell organelles and their functions — mitochondria, nucleus, cell membrane, ribosomes, chloroplasts, vacuole, cell wall" / Performance data: "End-of-topic quiz last Friday: class average 72%. Item-level breakdown: mitochondria 91% correct, nucleus 88%, cell membrane 74%, cell wall 69%, chloroplasts 65%, vacuole 58%, ribosomes 41%. Most common error: confusing ribosomes with rough ER, and confusing vacuole function (storage) with cell membrane function (selective permeability)" / Time horizon: "End-of-year exam in 14 weeks" / Available review slots: "Two 5-minute starter activities per week (Tuesday and Thursday)" / Student level: "Year 8, mixed ability"

---

Personalised Spacing Schedule: Cell Organelles

Content: 7 cell organelles and their functions (mitochondria, nucleus, cell membrane, ribosomes, chloroplasts, vacuole, cell wall) Performance snapshot: Class average 72%. Two items strong (mitochondria 91%, nucleus 88%). Two items moderate (cell membrane 74%, cell wall 69%). Three items weak (chloroplasts 65%, vacuole 58%, ribosomes 41%). Time horizon: 14 weeks until end-of-year exam Schedule basis: Items at highest forgetting risk (ribosomes, vacuole, chloroplasts) need immediate and frequent review. Items already strong (mitochondria, nucleus) need minimal review — just enough to prevent slow decay over 14 weeks. The schedule follows an expanding interval pattern: short gaps initially, lengthening as each item is successfully recalled.

Forgetting Risk Analysis

Item	Last Score	Forgetting Risk	Priority	Reasoning
Ribosomes	41%	Critical	1	Already below 50%. Without review this week, likely to drop to near-zero recall. The ribosome/rough ER confusion suggests a conceptual gap, not just weak memory.
Vacuole	58%	High	2	Just above chance. The confusion with cell membrane function suggests students have a fragile, easily-confused representation.
Chloroplasts	65%	High	3	Below class average. Likely being confused with mitochondria (both are energy-related). Needs review before the confusion hardens.
Cell wall	69%	Moderate	4	Just below average. Relatively straightforward concept but students may be confusing it with cell membrane.
Cell membrane	74%	Moderate	5	At class average. The selective permeability concept is abstract and likely to fade without reinforcement.
Nucleus	88%	Low	6	Strong recall. One review in weeks 3-4 should be sufficient to maintain.
Mitochondria	91%	Low	7	Strongest item. Review only if space allows, or combine with a comparison task (mitochondria vs. chloroplasts).

Review Schedule

Week 1:

• Tuesday: Ribosomes and vacuole (5 min). Format: retrieval practice — "Without looking at your notes, write down what ribosomes do and where they're found. Then do the same for vacuoles." Follow with immediate corrective feedback. These two are at critical risk and need retrieval within 4 days of the quiz.
• Thursday: Chloroplasts and cell wall (5 min). Format: comparison task — "How is a chloroplast different from a mitochondrion? How is a cell wall different from a cell membrane?" Comparison forces students to distinguish commonly confused pairs.

Week 2:

• Tuesday: Ribosomes, vacuole, and chloroplasts (5 min). Format: quick-fire quiz — 6 questions, 30 seconds each. Ribosomes and vacuole appear again because they were weakest; chloroplasts because the comparison task may not have fully resolved the confusion.
• Thursday: Cell membrane and cell wall (5 min). Format: application — "A plant cell is placed in salty water. Which organelles are involved in what happens next? What does each one do?" This reviews cell membrane and cell wall in a functional context.

Week 3:

• Tuesday: Ribosomes and vacuole (5 min). If Tuesday quiz in Week 2 showed improvement, use a harder format: "Explain why a cell with damaged ribosomes would eventually die." If still weak, return to basic retrieval.
• Thursday: All 7 organelles (5 min). Format: sketch and label from memory — "Draw a plant cell and label as many organelles as you can with their functions." First full review. Mitochondria and nucleus appear here for the first time since the original quiz — 2 weeks is the right interval for items at 88-91% recall with a 14-week retention horizon.

Week 4:

• Tuesday: Chloroplasts and any items that were wrong in Thursday's sketch (5 min). Adaptive — review what was forgotten.
• Thursday: Ribosomes and vacuole (5 min). Third review for the weakest items. If students are now above 80%, the interval can start expanding.

Weeks 5-6:

• One session per week reviewing ribosomes, vacuole, and chloroplasts (the originally weak items). Format: interleaved retrieval — mix organelle questions with other biology topics to increase difficulty. The interval has expanded from twice a week to once a week because these items should now be strengthening.
• One session per week reviewing cell membrane and cell wall, now at a longer interval.

Weeks 7-10:

• One session every 10 days reviewing all 7 organelles, interleaved with other topics. By this point, items should be stable enough for longer intervals. Use varied formats: multiple choice, short answer, diagram labelling, application questions.

Weeks 11-14 (pre-exam):

• One comprehensive review per week covering all 7 organelles alongside other revision topics. The interval is now approximately weekly — appropriate for items that have been successfully retrieved 5-6 times and need to be maintained for 3-4 more weeks.

Algorithm Explanation (Teacher-Friendly)

Here's the logic behind this schedule, in plain terms:

Why different items get different schedules: Think of each memory like a battery. The quiz told us which batteries are full (mitochondria — 91%), which are half-charged (cell membrane — 74%), and which are nearly dead (ribosomes — 41%). Nearly dead batteries need charging immediately and frequently. Full batteries just need a top-up every few weeks.

Why intervals expand: Every time a student successfully remembers something, the memory gets stronger — it holds its charge longer. So after a student correctly recalls "ribosomes make proteins" three times, they don't need to see it again for a week. After five successful recalls, they might not need it for two weeks. The schedule expands automatically as memories strengthen.

The 10-20% rule: Research (Cepeda et al., 2006) suggests a useful guideline — the gap between reviews should be roughly 10-20% of the total time until the knowledge is needed. Your exam is in 14 weeks (98 days), so the "maintenance" interval for well-known items is roughly every 10-20 days. For new or weak items, the interval starts much shorter and gradually works up to this.

Why comparison and application tasks appear: Retrieval practice ("write down what you remember") strengthens memory. But comparison tasks ("how is X different from Y?") and application tasks ("what happens when…") build UNDERSTANDING as well as memory. The schedule uses basic retrieval early on (when the memory is fragile) and more demanding formats later (when the memory is stable enough to handle harder questions).

Adaptation Rules

When a student gets an item RIGHT:

• If it's the first correct retrieval: schedule the next review in 3-4 days
• If it's the second correct retrieval: schedule in 7 days
• If it's the third correct retrieval: schedule in 14 days
• If it's the fourth or later: schedule in 21 days, then maintain at 2-3 week intervals

When a student gets an item WRONG:

• Reset the interval to 1-2 days — the item goes back to "critical"
• Use a different retrieval format (if they got it wrong on a quiz, try a comparison task)
• Do NOT simply re-teach the item — have the student attempt retrieval FIRST, even if they'll get it wrong, then provide corrective feedback. The failed retrieval attempt actually strengthens the subsequent learning (Kornell et al., 2009).

Implementation Tips

1. Keep it simple. You have two 5-minute slots per week. Don't try to track individual students' spacing schedules — that's what an AI system like Kaku is for. Instead, use the CLASS-LEVEL performance data to decide which items need review and when.

2. Use mini-whiteboards or exit tickets. The fastest way to do retrieval practice in 5 minutes is mini-whiteboards ("Show me: what do ribosomes do?") or exit tickets. These also give you quick performance data for adjusting the schedule.

3. Don't re-teach during review slots. The 5-minute starter is for RETRIEVAL, not re-teaching. If students can't remember, let them struggle briefly, then show the answer and move on. The struggle is productive — it strengthens the memory trace for next time.

4. Interleave after Week 4. Once items have been reviewed 2-3 times, start mixing them with other topics. Interleaving feels harder (students will feel less confident) but produces better long-term retention than blocked review.

5. The schedule is a starting point, not a script. If your Tuesday quiz shows that ribosomes have jumped to 85%, great — move them to a longer interval. If chloroplasts have dropped to 50%, move them back to a shorter interval. The schedule adapts to what students actually remember, not what you hope they remember.

---

Known Limitations

1. This skill designs and explains a spacing schedule — it does not execute it. Execution requires either a teacher implementing the schedule manually (feasible but effortful) or a system like Anki, Quizlet, or Kaku that can track individual item histories and calculate intervals automatically. The schedule above uses class-level data; truly personalised spacing requires individual-level tracking, which is impractical without technology.

2. The optimal spacing interval is an approximation. Cepeda et al.'s (2006) 10-20% guideline is a useful heuristic, not a precise formula. The actual optimal interval depends on material difficulty, encoding quality, individual differences, and the type of retrieval required. The schedule above is a principled starting point, not a mathematically optimised solution.

3. Spacing works for factual and conceptual knowledge but is less studied for procedural skills. The evidence base is strongest for vocabulary, factual recall, and conceptual understanding. For complex procedural skills (essay writing, mathematical problem-solving), the spacing effect still applies to the component knowledge, but the skill as a whole may need different practice structures.

4. Student compliance is the binding constraint. The best spacing algorithm in the world fails if students don't actually do the review. In a classroom, the teacher controls the review schedule. In homework or self-study contexts, many students will mass their practice (cramming) because it FEELS more effective, even though it isn't (Pashler et al., 2007). The schedule must be implemented, not just designed.

Intelligent Tutoring Dialogue Designer

What This Skill Does

Designs the dialogue logic for an AI tutoring interaction — when to ask a question, when to give a hint, when to explain, when to prompt for self-explanation, and when to stay silent. This is the hardest design problem in intelligent tutoring: too much intervention prevents productive struggle and creates dependency; too little leaves students stuck and frustrated. VanLehn (2011) showed that the effectiveness of tutoring (human or AI) depends on the quality of the step-level interaction — systems that engage students in active reasoning at each step dramatically outperform systems that simply present content and evaluate final answers. Chi et al. (2001) analysed what makes human tutoring effective and found, counter-intuitively, that the most effective tutors were NOT the ones who explained the most — they were the ones who asked the right questions and created opportunities for students to self-explain. The output includes a complete dialogue architecture (the phases and branching logic of the interaction), a library of dialogue moves (the specific things the tutor can say or do), decision rules (when to use each move based on student responses), and a worked example showing the system in action. AI is specifically valuable here because it can sustain one-to-one dialogue at scale — but the dialogue must be deliberately designed or the AI will default to lecturing, which is the least effective tutoring strategy.

Evidence Foundation

VanLehn (2011) conducted a comprehensive meta-analysis comparing human tutoring, intelligent tutoring systems, and no-tutoring conditions. He found that the critical factor was not WHO was tutoring but HOW. "Inner loop" systems — those that provided feedback and scaffolding at each problem-solving step — achieved effect sizes of 0.76, nearly matching human tutors (0.79). "Outer loop" systems — those that only evaluated final answers — achieved much lower effect sizes (0.31). The implication for dialogue design is clear: the system must engage with the student's reasoning process, not just their final answer. Chi et al. (2001) conducted detailed analysis of effective human tutoring dialogues and identified a surprising finding: the most effective tutors did NOT give the best explanations. Instead, they used a pattern of "elicit, then explain" — first prompting the student to attempt an explanation, then building on whatever the student produced. This finding directly contradicts the intuitive assumption that good tutoring is about clear explanation. The reason: when a tutor explains, the student passively receives. When a tutor prompts and the student attempts to explain, the student actively constructs understanding (Chi & Wylie, 2014 — the ICAP framework). Graesser et al. (2005) developed AutoTutor, one of the most extensively researched intelligent tutoring systems, which uses a "mixed-initiative dialogue" approach. AutoTutor asks questions, evaluates student responses, gives feedback, and prompts for elaboration — maintaining a conversational exchange rather than a lecture. Their research identified five key dialogue moves: pumps ("Tell me more"), prompts (asking for specific information), hints (pointing toward the answer without giving it), assertions (providing information), and corrections (directly addressing errors). Koedinger & Aleven (2007) articulated the "assistance dilemma" — the fundamental tension in tutoring design. Too much assistance (explaining everything, giving hints too quickly) produces shallow learning: the student completes the task but doesn't understand why. Too little assistance (never intervening, making students struggle endlessly) produces frustration and abandonment. The optimal tutoring strategy navigates between these extremes, providing the MINIMUM assistance necessary for the student to make progress.

Input Schema

The teacher must provide:

• Learning objective: What the student should master. e.g. "Understanding why heavier objects do NOT fall faster than lighter objects (Newton's vs. Aristotelian physics)" / "Being able to identify the main argument in a non-fiction text and distinguish it from supporting evidence" / "Understanding that multiplying by a fraction less than 1 makes a number smaller, not bigger"
• Anticipated difficulties: Where students struggle. e.g. "Students have the Aristotelian intuition that heavier = faster. They cite everyday experience (dropping a feather vs. a ball) as 'proof.' They struggle to distinguish air resistance from gravitational acceleration" / "Students confuse the topic of a text with the argument. They identify facts rather than claims. They struggle to distinguish what the author believes from what the author reports" / "Students apply the 'multiplication makes bigger' rule from whole numbers and are confused when ½ × 6 = 3"

Optional (injected by context engine if available):

• Student level: Year group and proficiency
• Subject area: The curriculum subject
• Interaction length: How long the interaction should last
• Student model: Known information about the specific student
• System capabilities: What the AI system can do

Prompt

You are an expert in tutoring dialogue design, with deep knowledge of VanLehn's (2011) meta-analysis of ITS effectiveness, Chi et al.'s (2001) research on effective tutoring dialogues, Graesser et al.'s (2005) AutoTutor research, Chi & Wylie's (2014) ICAP framework, and Koedinger & Aleven's (2007) assistance dilemma. You understand that the quality of tutoring depends not on what the tutor KNOWS but on what the tutor DOES — specifically, how the tutor manages the dialogue to maximise the student's active cognitive engagement.

CRITICAL PRINCIPLES:
- **Elicit before explain.** The most common mistake in tutoring (human and AI) is explaining too much, too early. The first move should almost always be a QUESTION, not an explanation. Even when the student is clearly wrong, the first move is "Can you explain your thinking?" — not "Actually, the correct answer is..." Chi et al. (2001) showed that tutoring effectiveness correlates with how much the STUDENT talks, not how much the tutor talks.
- **Navigate the assistance dilemma.** Koedinger & Aleven (2007): too much help → shallow learning; too little help → frustration. The dialogue should provide the MINIMUM assistance necessary for progress. Start with the lightest possible intervention (a question, a pause, a pump) and escalate ONLY when lighter moves fail.
- **Use mixed-initiative dialogue.** Graesser et al. (2005): effective tutoring is a CONVERSATION, not a lecture or a quiz. The tutor sometimes leads (asking questions) and sometimes follows (responding to student questions). The student should have genuine agency in the dialogue — they can ask questions, express confusion, change the topic, or disagree.
- **Silence is a move.** One of the most powerful tutoring moves is to say nothing. After asking a question, WAIT. Students need time to think. If the system responds instantly to every student action, it trains the student to be passive. Deliberate silence (with visible thinking time) creates productive struggle.
- **Design for misconceptions, not just errors.** An error is getting the answer wrong. A misconception is having a wrong mental model. The dialogue must distinguish between these: errors are corrected with feedback; misconceptions are addressed by creating cognitive conflict — presenting evidence that contradicts the student's mental model and prompting them to reconcile the conflict.

Your task is to design the dialogue logic for:

**Learning objective:** {{learning_objective}}
**Anticipated difficulties:** {{anticipated_difficulties}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Student level:** {{student_level}} — if not provided, design for a general secondary school context.
**Subject area:** {{subject_area}} — if not provided, infer from the learning objective.
**Interaction length:** {{interaction_length}} — if not provided, design for a 10-15 minute interaction.
**Student model:** {{student_model}} — if not provided, assume a typical student who may hold the anticipated misconceptions.
**System capabilities:** {{system_capabilities}} — if not provided, design for a text-based AI system that can also display images and worked examples.

Return your output in this exact format:

## Tutoring Dialogue Design: [Learning Objective]

**Learning objective:** [What the student should master]
**Anticipated difficulties:** [Where students struggle]
**Interaction model:** [How the dialogue is structured]

### Dialogue Architecture

[The overall structure of the interaction — phases, transitions, and branching logic. Include a visual flow if helpful.]

**Phase 1: [Name]** — [Purpose and key moves]
**Phase 2: [Name]** — [Purpose and key moves]
**Phase 3: [Name]** — [Purpose and key moves]

### Dialogue Move Library

[The specific moves available to the tutor at each point, with examples of each]

| Move Type | When to Use | Example |
|---|---|---|
| [Move] | [Condition] | [Exact text] |

### Decision Rules

[The rules that govern which move to use, based on student responses. Format as IF-THEN rules.]

### Example Dialogue

[A complete, realistic dialogue showing the system in action with a student who holds the anticipated misconception. Show BOTH effective and challenging moments — including what happens when the student is confused, resistant, or wrong.]

**Tutor:** [What the system says]
**Student:** [A realistic student response]
**Tutor:** [Response, with annotation explaining WHY this move was chosen]
...

### Design Rationale

[Why the dialogue is structured this way — what research principles it follows and what alternatives were considered.]

**Self-check before returning output:** Verify that (a) the dialogue elicits before explaining, (b) the assistance dilemma is navigated (minimum necessary help), (c) silence is used as a deliberate move, (d) misconceptions are addressed through cognitive conflict not just correction, and (e) the student talks more than the tutor.

Example Output

Scenario: Learning objective: "Understanding why heavier objects do NOT fall faster than lighter objects — the distinction between gravitational acceleration (same for all objects) and air resistance (different for different objects)" / Anticipated difficulties: "Students have strong Aristotelian intuition that heavier = faster. They cite everyday experience (dropping a feather vs. a ball) as 'proof.' They struggle to separate the concept of air resistance from gravitational acceleration. Even after being told the correct answer, many revert to the misconception within weeks" / Student level: "Year 9, mixed ability science class" / Subject area: "Physics — forces and motion"

---

Tutoring Dialogue Design: Gravitational Acceleration vs. Air Resistance

Learning objective: Understanding why heavier objects do NOT fall faster than lighter objects (same gravitational acceleration, different air resistance) Anticipated difficulties: Strong Aristotelian intuition (heavier = faster), everyday experience cited as proof, air resistance confused with gravity Interaction model: Mixed-initiative dialogue across three phases — elicit the misconception, create cognitive conflict, build the correct model

Dialogue Architecture

Phase 1: Elicit (3-4 minutes) — Surface the student's existing belief WITHOUT correcting it. The goal is to get the student to articulate their mental model clearly and commit to a prediction. This is essential because (a) you cannot address a misconception you haven't surfaced, and (b) making a clear prediction creates the conditions for productive cognitive conflict when the prediction is contradicted.

Phase 2: Conflict (4-5 minutes) — Present a scenario that contradicts the student's prediction. Do NOT explain why the prediction was wrong — ask the student to explain the contradiction. This phase is the heart of the dialogue. The student must experience genuine cognitive conflict: "My prediction was wrong, but I was SURE it was right. Why?" The tutor's role is to maintain the conflict (not resolve it too quickly) and to guide the student toward the key distinction (gravity vs. air resistance).

Phase 3: Construct (4-5 minutes) — Help the student build the correct mental model. Use self-explanation prompts to ensure the student can articulate the distinction between gravitational acceleration (same for all objects) and air resistance (depends on shape and surface area, not mass). End with a transfer question to test whether the model is genuinely understood or just memorised.

Dialogue Move Library

Move Type	When to Use	Example
Prediction request	Phase 1 — to surface the misconception	"If I dropped a bowling ball and a tennis ball from the same height at the same time, which would hit the ground first? What's your prediction?"
Commitment probe	After prediction — to strengthen commitment	"How confident are you? And can you explain WHY you think that?"
Pump	When student gives a short answer	"Tell me more about that." / "What makes you think so?"
Conflict scenario	Phase 2 — to create cognitive dissonance	"Now imagine both objects are in a room with no air — a vacuum. Same prediction?"
Silence	After any question — give thinking time	[Wait 5-8 seconds before responding. If the system is text-based, display "..." or "Take your time."]
Redirect	When student goes off-topic	"That's interesting — let's come back to that. Right now, I want to focus on: why did the vacuum change the result?"
Socratic question	Phase 2/3 — to guide reasoning	"So if gravity pulls both the same, what's the ONLY thing that could make one fall slower than the other?"
Self-explanation prompt	Phase 3 — to build understanding	"In your own words, explain why a feather falls slowly in air but fast in a vacuum."
Minimal correction	When student makes a factual error	"Actually, gravity accelerates all objects at the same rate — 9.8 m/s². But your question about air is really important. Keep going with that thought."
Transfer question	End of Phase 3 — to test understanding	"A parachutist and a skydiver without a parachute jump from the same plane. Using what we've discussed, explain what happens and why."

Decision Rules

IF the student predicts "heavier falls faster" with high confidence: → THEN proceed to Phase 2 with the vacuum thought experiment. The misconception is present and committed — ideal for cognitive conflict.

IF the student predicts "heavier falls faster" but seems uncertain: → THEN use a commitment probe: "Interesting — can you explain your reasoning?" This strengthens their commitment to the prediction, which makes the conflict more productive.

IF the student gives the correct answer immediately: → THEN probe for understanding: "That's right. But most people think heavier objects fall faster. Can you explain WHY they're wrong — what makes people think that?" This tests whether the student genuinely understands or has just memorised the fact.

IF the student responds to the vacuum scenario correctly: → THEN use a self-explanation prompt: "So you're saying they'd fall at the same speed in a vacuum. What does that tell you about GRAVITY specifically?" Move to Phase 3.

IF the student responds to the vacuum scenario but is confused: → THEN use a Socratic question: "You said they'd fall at the same speed in a vacuum but different speeds in air. What's DIFFERENT about those two situations?" Guide them toward identifying air resistance.

IF the student is silent for more than 10 seconds: → THEN offer a lighter version of the question: "Let me put it differently. In a vacuum, there's no air. In a room, there IS air. What effect does air have on a falling object?"

IF the student asks a question: → THEN answer it briefly and redirect to the main line of inquiry. Do not lecture. "Good question — yes, gravity is the same everywhere near Earth's surface. Now, given that, why does a feather fall differently from a ball?"

IF the student says "I don't know" or "I'm stuck": → THEN do NOT explain. Offer a scaffold: "Let's think about it step by step. First: in a vacuum, both objects fall at the same speed. Do you agree?" Build from what they DO know.

Example Dialogue

Tutor: I want you to imagine something. You're standing on the roof of the school holding a bowling ball in one hand and a tennis ball in the other. You drop them both at exactly the same time from the same height. Which one hits the ground first?

Student: The bowling ball, obviously. It's heavier.

Tutor: You sound quite sure. Can you explain why you think the heavier one falls faster? [Commitment probe — getting the student to articulate and commit to their reasoning]

Student: Because it's heavier? Like, gravity pulls it more because it has more mass. So it accelerates faster.

Tutor: OK, so your reasoning is: more mass → more gravitational pull → falls faster. Let me write that down: "Heavier objects experience more gravitational pull, so they accelerate faster and hit the ground first." Is that a fair summary of your position? [Explicit commitment — the student's prediction is now on the record, setting up the conflict]

Student: Yeah, that's what I think.

Tutor: Good. Now I want you to imagine a different version of the same experiment. Same bowling ball, same tennis ball, same height. But this time, you're in a room where ALL the air has been pumped out — a perfect vacuum. Nothing but you and the two balls. You drop them both. What happens? [Conflict scenario — introducing the vacuum condition]

Student: ...I think they'd still fall at different speeds? The bowling ball is still heavier.

Tutor: ... [Silence — 6 seconds. Letting the student sit with their prediction.]

Tutor: Here's something that might surprise you. We've actually done this experiment — many times. In a vacuum, the bowling ball and the tennis ball hit the ground at EXACTLY the same time. Every single time. What do you make of that? [Presenting the disconfirming evidence without explaining it]

Student: Wait, really? That doesn't make sense. The bowling ball is way heavier.

Tutor: I know it feels like it shouldn't happen. But it does. So here's my question: you predicted they'd fall at different speeds, but in a vacuum they don't. What went wrong with the prediction? [Socratic question — asking the student to diagnose their own error]

Student: Um... maybe gravity pulls them the same even though one is heavier?

Tutor: That's really interesting. Keep going with that thought. If gravity pulls them the same in a vacuum, why do they fall at different speeds in NORMAL air? [Pump + redirect — the student is close to the key insight, guiding them toward air resistance]

Student: Oh... is it because of the air? Like, the air slows down the lighter one more?

Tutor: Say more about that. Why would air slow down one more than the other? [Pump — maintaining the student's cognitive engagement, not jumping to explain]

Student: Because the tennis ball is lighter so the air resistance affects it more? Like, it can't push through the air as easily?

Tutor: You're onto something really important. Let me check your understanding. In your own words, explain: why does a feather fall slowly in air but at the same speed as a hammer in a vacuum? [Self-explanation prompt — testing whether the student has built a coherent model]

Student: Because in a vacuum there's no air resistance, so only gravity matters, and gravity accelerates everything the same. But in air, the feather has way more air resistance compared to its weight, so it falls slower. The hammer barely notices the air.

Tutor: That's an excellent explanation. You've identified the key distinction: gravitational acceleration is the same for all objects, but air resistance is different depending on shape, surface area, and mass. Now let me test that understanding with a new situation. A parachutist and a skydiver WITHOUT a parachute jump from the same plane at the same time. Using what we've just discussed, predict what happens and explain why. [Transfer question — testing whether the model generalises]

Design Rationale

The dialogue follows a deliberate three-phase structure based on conceptual change research:

1. Elicit before correct. The tutor spends the first phase surfacing and committing the student to their Aristotelian belief. This is counterintuitive — many teachers would immediately say "Actually, they fall at the same speed." But Chi et al. (2001) showed that direct correction of misconceptions is largely ineffective. The student needs to feel the conflict between their prediction and reality.

2. Cognitive conflict, not correction. Phase 2 presents a fact (same speed in vacuum) that contradicts the student's prediction, then asks the STUDENT to explain the discrepancy. The tutor does not explain — the student must do the cognitive work of reconciling their prediction with the evidence. This is the highest-engagement dialogue pattern in the ICAP framework (Chi & Wylie, 2014).

3. Student-generated explanation. By Phase 3, the student has arrived at the correct explanation ("air resistance, not gravity, causes the difference") through their own reasoning. The tutor's self-explanation prompt checks this understanding, and the transfer question tests whether it generalises. The student talks more than the tutor throughout.

4. Minimum assistance. The tutor uses the lightest possible move at each point: silence, pumps, and questions before hints or explanations. The only direct assertion the tutor makes is the factual statement that objects fall at the same speed in a vacuum — everything else is drawn from the student through questioning.

---

Known Limitations

1. This skill designs dialogue LOGIC, not dialogue IMPLEMENTATION. Converting the decision rules and move library into a working AI system requires engineering (state tracking, natural language understanding, response generation) that goes far beyond this skill's output. The design is a blueprint, not a deployable system.

2. The dialogue assumes one misconception at a time. Real students often hold multiple, interacting misconceptions simultaneously. A student who confuses mass with weight AND doesn't understand air resistance presents a more complex tutoring challenge than the single-misconception dialogue above. Multi-misconception dialogue design is a research frontier, not a solved problem.

3. Mixed-initiative dialogue is hard for current AI systems. The dialogue above assumes the AI can understand student responses, detect misconceptions, and select appropriate moves in real time. Current LLMs can approximate this in many cases, but they lack the reliable state tracking and misconception detection of purpose-built ITS systems like AutoTutor. The dialogue may need to be simplified for deployment on current systems.

4. The evidence base is primarily from STEM domains. VanLehn (2011), Graesser et al. (2005), and Koedinger & Aleven (2007) conducted their research primarily in mathematics and science. The dialogue principles transfer to other domains (the "elicit before explain" pattern works in humanities too), but the specific patterns of misconception and conflict may differ. Tutoring dialogue in essay writing or historical analysis looks different from tutoring dialogue in physics, even though the underlying principles are the same.

5. Silence is difficult in text-based AI interactions. The dialogue prescribes deliberate silence as a tutoring move, but in a text-based interface, silence can be indistinguishable from system failure. Implementing "productive silence" in a chatbot requires explicit design — a visible timer, a "Take your time" message, or a deliberate delay before the next prompt.

Learning Analytics Interpretation Guide

What This Skill Does

Guides a teacher through interpreting a specific learning dataset — assessment results, engagement metrics, study behaviour patterns, or any other quantitative or qualitative data about student learning — to identify actionable patterns and inform specific teaching decisions. The critical insight from Wiliam (2011) and Mandinach & Gummer (2016) is that learning analytics is ONLY useful if it changes teacher decisions. A dashboard full of colourful graphs is worthless if the teacher doesn't know what to DO differently as a result. This skill bridges the gap between data and action: it takes raw data, identifies the patterns that matter, explains them in plain language, recommends specific teaching responses, and — critically — flags what the data does NOT show and the interpretive traps the teacher should avoid. AI is specifically valuable here because interpreting learning data requires simultaneously considering multiple variables (individual vs. group patterns, prior performance, assessment validity, possible confounds) — a cognitive task that is difficult for a teacher reviewing data at 8pm after a full teaching day, but straightforward for a well-designed AI system.

Evidence Foundation

Siemens & Long (2011) articulated the foundational vision for learning analytics: using data generated by learners to understand and optimise learning. They distinguished between academic analytics (institutional-level data for strategic decisions) and learning analytics (course/student-level data for teaching decisions). The key insight: the value of analytics is not in the data itself but in the decisions it enables. Bienkowski et al. (2012) produced a comprehensive US Department of Education report on educational data mining and learning analytics, reviewing the evidence base and identifying key applications: early warning systems (identifying at-risk students), adaptive learning (adjusting content to individual performance), and formative feedback (informing day-to-day teaching decisions). They found that the most effective applications were those that provided teachers with actionable information, not raw data dumps. Wiliam (2011) argued that data use in education should be fundamentally FORMATIVE — the purpose of collecting data is to adjust teaching, not to label students. He identified five key formative assessment strategies, all of which depend on effective data interpretation: clarifying learning intentions, engineering effective discussions, providing feedback that moves learners forward, activating students as instructional resources for one another, and activating students as owners of their own learning. Mandinach & Gummer (2016) studied teacher data literacy and found that most teachers lack training in data interpretation. The most common errors: confusing correlation with causation, over-interpreting small samples, ignoring measurement error, and focusing on averages while missing important subgroup patterns. They argued that data literacy is a core teaching competence that is rarely taught in initial teacher education. Wise (2014) found that simply giving students access to their own learning analytics did not improve learning — students needed structured guidance on how to interpret and act on the data. The same principle applies to teachers.

Input Schema

The teacher must provide:

• Dataset description: What data is available. e.g. "Year 10 mock exam results: class of 28, scores range from 23% to 91%, mean 58%, median 54%. Question-level data available: Q1 (recall) 78% correct, Q2 (application) 45% correct, Q3 (analysis) 31% correct, Q4 (evaluation) 28% correct" / "Kaku engagement data for Year 7 maths: 24 students completed homework last week, 6 didn't. Of those who completed it, average time spent was 18 minutes. Three students spent over 40 minutes. Two students completed in under 5 minutes with 100% accuracy" / "Reading log data: 30 students, 3 weeks of daily reading minutes. Class average dropped from 22 min/day in week 1 to 14 min/day in week 3"
• Decision context: What the teacher needs to decide. e.g. "I need to plan the next two weeks of revision lessons — what should I focus on?" / "I'm deciding whether to set more challenging homework or to go back and reteach the basics" / "I'm concerned about the reading decline and need to decide whether to intervene or whether this is normal variation"

Optional (injected by context engine if available):

• Student level: Year group and proficiency
• Subject area: The curriculum subject
• Comparison data: Baselines or benchmarks
• Time constraints: How quickly action is needed

Prompt

You are an expert in learning analytics interpretation, with deep knowledge of Siemens & Long's (2011) learning analytics framework, Bienkowski et al.'s (2012) DoE report on educational data mining, Wiliam's (2011) formative assessment principles, Mandinach & Gummer's (2016) research on teacher data literacy, and Wise's (2014) work on structured analytics interpretation. You understand that the purpose of learning data is to inform TEACHING DECISIONS — not to produce graphs or label students. You also understand the common interpretive traps: confusing correlation with causation, over-interpreting small samples, ignoring measurement error, and focusing on averages while missing critical subgroup patterns.

CRITICAL PRINCIPLES:
- **Start with the decision, not the data.** The teacher has a specific decision to make. The data interpretation should be organised around THAT decision, not around every possible pattern in the data. A data dump is not an interpretation.
- **Distinguish signal from noise.** Small differences in scores, engagement times, or completion rates may be meaningful or they may be random variation. Before recommending action based on a pattern, consider: is this difference large enough to be meaningful? Could it be explained by measurement error, bad questions, or chance? Be honest about uncertainty.
- **Look for subgroup patterns, not just averages.** A class average of 58% could mean most students scored around 58%, OR it could mean half the class scored 80%+ and half scored below 40%. These are completely different situations requiring completely different responses. Always examine the DISTRIBUTION, not just the central tendency.
- **Prioritise by actionability.** Not all patterns are equally actionable. "Q3 was hard" is less actionable than "Q3 was hard because students couldn't apply the concept of opportunity cost to a novel scenario — they could define it (Q1) but not use it." The interpretation should connect data patterns to specific teaching actions.
- **Flag what the data does NOT show.** Every dataset has blind spots. Assessment data shows what students PRODUCED, not what they UNDERSTOOD. Engagement data shows time spent, not learning achieved. Completion data shows who finished, not who benefited. The teacher needs to know the limits of their data, not just the patterns.

Your task is to interpret this data and guide the teacher's decision:

**Dataset description:** {{dataset_description}}
**Decision context:** {{decision_context}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Student level:** {{student_level}} — if not provided, infer from the data.
**Subject area:** {{subject_area}} — if not provided, infer from the data.
**Comparison data:** {{comparison_data}} — if not provided, note what comparisons would be useful.
**Time constraints:** {{time_constraints}} — if not provided, assume the teacher needs guidance for this week's planning.

Return your output in this exact format:

## Data Interpretation: [Brief Description]

**Data:** [Summary of what data is available]
**Decision:** [What the teacher needs to decide]
**Headline finding:** [One sentence — the most important pattern in the data for this decision]

### What the Data Shows

[Clear, jargon-free interpretation of the key patterns. Use the data to tell a story. Organise by relevance to the teacher's decision, not by data type.]

### Actionable Patterns

[Ranked list of patterns that suggest specific teaching actions. For each:]

**Pattern [N]: [Name]**
- **What the data shows:** [The specific numbers]
- **What it probably means:** [The most likely interpretation]
- **Confidence level:** [How confident you are — high/moderate/low — and why]
- **Suggested action:** [What to do about it]

### Recommended Actions

[Concrete, prioritised teaching actions linked to the data patterns above. What to do this week, what can wait.]

### What the Data Does NOT Show

[Critical caveats. What should the teacher NOT conclude from this data? What alternative explanations exist? What additional data would be needed to confirm the interpretation?]

### Quick Reference

| Pattern | Action | Priority | Confidence |
|---|---|---|---|
| [Pattern] | [Action] | [High/Medium/Low] | [High/Moderate/Low] |

**Self-check before returning output:** Verify that (a) the interpretation is organised around the teacher's decision, (b) signal is distinguished from noise, (c) subgroup patterns are examined, (d) actions are specific and prioritised, and (e) limitations of the data are honestly flagged.

Example Output

Scenario: Dataset description: "Year 10 mock exam results — GCSE Geography Paper 1. Class of 28 students. Scores range from 23% to 91%. Mean: 58%. Median: 54%. Standard deviation: 18%. Question-level data: Q1 (define key terms — 4 marks) 78% correct. Q2 (describe patterns from a map — 6 marks) 62% correct. Q3 (explain formation of landform using a case study — 8 marks) 31% correct. Q4 (evaluate the effectiveness of a management strategy — 12 marks) 28% correct. Student-level distribution: 6 students scored 75%+, 10 students scored 50-74%, 8 students scored 35-49%, 4 students scored below 35%." / Decision context: "I have 6 lessons before the real GCSE exam. I need to decide what to prioritise in revision — should I reteach the content for Q3/Q4, practise exam technique, or focus on the weakest students?" / Subject area: "GCSE Geography" / Comparison data: "Last year's class averaged 62% on the same paper. National average for this paper is typically around 55%"

---

Data Interpretation: Year 10 GCSE Geography Mock Exam

Data: 28 students, Paper 1 mock exam, scores 23-91%, mean 58%, question-level and student-level breakdowns Decision: How to use 6 remaining lessons before the real exam — content reteaching, exam technique, or targeted student support Headline finding: The class can recall and describe (Q1-Q2 adequate) but cannot explain or evaluate (Q3-Q4 critically weak) — this is an exam SKILLS gap, not primarily a knowledge gap.

What the Data Shows

The big picture: Your class average (58%) is slightly below last year's class (62%) but above the typical national average (55%). This suggests the class is broadly on track but has specific, addressable weaknesses.

The critical pattern is in the question types, not the content: Q1 (recall — define terms) at 78% and Q2 (describe patterns from a map) at 62% tell you that most students KNOW the content. They can define the terms and read the data. But Q3 (explain formation using a case study) at 31% and Q4 (evaluate a management strategy) at 28% tell you they cannot DO anything with that knowledge. The drop from Q2 (62%) to Q3 (31%) is steep — this is where most marks are being lost, and it's where most marks are available (Q3 + Q4 = 20 of the 30 marks).

The student distribution matters: You don't have a uniform class. You have three distinct groups:

• Group A (6 students, 75%+): These students are performing well across all question types. They need refinement, not reteaching.
• Group B (10 students, 50-74%): These students know the content (Q1/Q2) but struggle with application (Q3/Q4). They are your BIGGEST opportunity for improvement — they have the knowledge base, they just need to learn how to deploy it in exam conditions.
• Group C (12 students, below 50%): These students have mixed issues. The 8 students at 35-49% probably have the same skills gap as Group B, plus some knowledge gaps. The 4 students below 35% may have fundamental knowledge gaps that 6 lessons cannot fully address.

Actionable Patterns

Pattern 1: Explain/Evaluate Skills Gap

• What the data shows: 78% on recall, 62% on describe, 31% on explain, 28% on evaluate
• What it probably means: Students can recall content but cannot structure extended answers. They likely lack a clear method for "explain" questions (e.g., Point-Evidence-Explain chains) and "evaluate" questions (e.g., For-Against-Judgement structures). This is an exam technique issue, not a knowledge issue.
• Confidence level: High — the pattern is consistent and the drop between Q2 and Q3 is too steep to be explained by content gaps alone. If students didn't know the content, Q1 would also be low.
• Suggested action: Dedicate 3-4 of your 6 lessons to explain/evaluate exam technique. Use model answers, live modelling ("watch me write a Q3 answer"), and scaffolded practice.

Pattern 2: Case Study Application Weakness

• What the data shows: Q3 (explain using a case study) at 31% — the lowest of the "knowledge" questions (Q1-Q3)
• What it probably means: Students either don't have case study details memorised OR they can't integrate case study details into an explanation. Review a sample of student scripts to check: are they attempting the case study and getting it wrong, or are they skipping the case study references entirely?
• Confidence level: Moderate — you need to look at the actual scripts to distinguish "don't know the case study" from "don't know how to use the case study in an explanation."
• Suggested action: Quick diagnostic: give students 2 minutes to write down everything they remember about the relevant case study. If they can recall facts, the issue is application technique. If they can't, the issue is knowledge and they need a rapid case study revision resource.

Pattern 3: Bimodal Distribution

• What the data shows: Mean 58%, median 54%, range 23-91%, with distinct clusters at 75%+, 50-74%, and below 50%
• What it probably means: The class is not a single group — it's at least two or three groups with different needs. A single revision strategy will not work for all students.
• Confidence level: High — the distribution is clear from the data.
• Suggested action: Differentiate your revision lessons. Group A can work on stretch and refinement (higher-order evaluation). Group B needs structured exam technique for Q3/Q4 questions. Group C (especially the 4 below 35%) needs a combination of rapid content consolidation and basic exam structure.

Recommended Actions

Lesson 1 (this week): Whole-class model answer session. Take Q3 from the mock and live-model a high-scoring answer. Show the structure explicitly: "Here's how I link the case study to the explanation." Then give students a similar Q3 to attempt in 10 minutes. This addresses Pattern 1 immediately.

Lessons 2-3: Scaffolded practice on explain and evaluate questions. Provide sentence starters and paragraph frameworks for Group B/C. Group A works independently on full-mark model answers. Quick case study recall quiz at the start of each lesson (Pattern 2).

Lesson 4: Focused evaluate question practice. This is where the most marks are available (12 marks) and where performance was weakest (28%). Teach a clear evaluate structure (for/against/judgement with evidence) and practise with exam-style questions.

Lesson 5: Timed practice under exam conditions — one full Paper 1 section. This builds stamina and time management, which are often hidden causes of poor Q3/Q4 performance (students run out of time because they spend too long on Q1/Q2).

Lesson 6: Review timed practice, address remaining gaps, and consolidate exam technique. Personalised revision lists for each student based on their mock and timed practice performance.

What the Data Does NOT Show

1. The data does not show WHY Q3/Q4 scores are low. The two most likely explanations are (a) students lack exam technique for extended answers and (b) students lack the case study knowledge to answer Q3. These require DIFFERENT interventions. Look at the actual scripts before committing to a strategy.

2. The data does not account for student effort and test conditions. A mock exam is not a real exam. Some students may have underperformed because they didn't revise seriously, were tired, or didn't feel the stakes were real. The 4 students below 35% may include students who gave up partway through — their knowledge may be better than their scores suggest.

3. The comparison with last year is imprecise. Your class averaged 58% vs. last year's 62%. But these are different students taking the same paper — the difference may reflect cohort ability, not teaching quality. Do not use this comparison to evaluate your own effectiveness.

4. Question-level data doesn't reveal individual misconceptions. Q3 at 31% tells you THAT students struggled, not WHERE in the question they went wrong. Review a sample of 5-6 scripts across ability levels to identify the specific sticking points.

Quick Reference

Pattern	Action	Priority	Confidence
Explain/evaluate skills gap (Q3-Q4)	Model answers + scaffolded practice	High	High
Case study application weakness	Quick diagnostic quiz, then targeted revision	High	Moderate
Bimodal distribution	Differentiated revision plan by group	Medium	High
Low scores in bottom 4 students	Individual check-in, targeted support	Medium	Moderate

---

Known Limitations

1. This skill interprets data — it does not collect or validate it. The quality of the interpretation depends entirely on the quality of the data provided. If the assessment was poorly designed (ambiguous questions, misaligned mark schemes), the data will be misleading and the interpretation will be misleading. Garbage in, garbage out.

2. Learning analytics can identify patterns but rarely proves causation. When this skill says "students probably lack exam technique," that is an inference, not a certainty. Alternative explanations always exist. The teacher should treat the interpretation as a starting hypothesis to investigate, not a confirmed diagnosis.

3. The evidence base for learning analytics in K-12 is less developed than in higher education. Siemens & Long (2011) and much of the learning analytics literature focuses on university-level data (LMS logs, course completion rates). The principles transfer to school contexts, but the specific patterns and benchmarks may differ. Mandinach & Gummer (2016) note that teachers need domain-specific data literacy, not generic statistical skills.

4. Analytics can inadvertently narrow the curriculum. If teachers consistently use assessment data to drive revision, there is a risk that teaching becomes focused on assessed outcomes at the expense of broader learning. Wiliam (2011) argues that formative data should inform teaching, not define it. The data shows how students performed on THIS test — not what they need to learn most.

Metacognitive Monitoring in AI Contexts

What This Skill Does

Analyses how AI tool use in a specific learning context might distort students' metacognitive monitoring — their ability to accurately assess what they know and don't know — and designs interventions to maintain metacognitive accuracy. This is one of the most urgent challenges in AI-enabled education. When a student uses an AI tool to complete work, they may experience a fluency illusion: the work looks good, the answers are correct, the text is fluent — and the student concludes "I understand this." But the STUDENT didn't do the cognitive work; the AI did. The student's sense of understanding is calibrated to the PRODUCT (which is good) rather than to their OWN knowledge (which may be unchanged). Bjork et al. (2013) showed that learners are systematically poor at judging their own learning — they confuse familiarity with understanding, and fluent performance with durable knowledge. AI tools dramatically amplify this miscalibration because they produce fluent, correct output that the student may mistake for evidence of their own competence. The output includes a metacognitive diagnosis (how AI use distorts self-assessment in this specific context), monitoring interventions (strategies to improve metacognitive accuracy), AI usage guidelines (when to use and when to restrict AI), and assessment alignment (ensuring tests measure student knowledge, not AI-assisted performance).

Evidence Foundation

Winne & Hadwin (1998) developed the most comprehensive model of self-regulated learning (SRL), which places metacognitive monitoring at its centre. Their model describes a cycle: the learner sets goals, applies strategies, monitors whether the strategies are working, and adjusts. Effective learning depends critically on the MONITORING stage — the learner's ability to accurately judge whether they are understanding the material. When monitoring is inaccurate (the learner thinks they understand when they don't), the entire self-regulation cycle breaks down: they stop studying too early, choose inappropriate strategies, and are surprised by poor assessment results. Thiede et al. (2003) showed that metacomprehension accuracy (the correlation between judged and actual understanding) is typically very low — around r = 0.27. However, they found that certain activities dramatically improve accuracy: delayed summary writing, keyword generation, and any task that forces the learner to generate from memory rather than recognise from the text. The key principle: metacognitive accuracy improves when the monitoring task requires RETRIEVAL, not just recognition. Dunning et al. (2003) documented the Dunning-Kruger effect: the least competent individuals are the MOST overconfident in their abilities, because they lack the knowledge needed to recognise their own incompetence. In AI contexts, this effect may be amplified: a student who doesn't understand a concept cannot distinguish their own (poor) understanding from the AI's (excellent) output. Bjork et al. (2013) reviewed the psychology of self-regulated learning and identified several "illusions of competence" — conditions where learners feel they've learned more than they actually have. These include: familiarity (having seen something before feels like understanding it), fluency (material that's easy to process feels like it's well-learned), and performance (doing well now feels like permanent learning). AI tools can trigger all three illusions simultaneously: the AI-produced output is familiar (the student saw it being generated), fluent (LLMs produce polished text), and high-performing (the answers are correct). Kazemitabaar et al. (2023) studied how AI code generators (like Copilot) affect novice programming learners and found that while AI-assisted students completed tasks faster and with fewer errors, they showed weaker understanding on subsequent tasks without AI support. The students had learned to use the AI, not to program. This is a direct empirical demonstration of the metacognitive risk: AI assistance produced the ILLUSION of learning without the REALITY of learning.

Input Schema

The teacher must provide:

• AI learning context: How students are using AI. e.g. "Year 12 students use ChatGPT to help write A-level English Literature essays. They paste in essay questions and use the AI output as a starting point, then edit and refine" / "Year 9 students use an AI maths tutor that solves equations step-by-step when they get stuck. They can ask for help at any point" / "Year 10 students use AI to generate revision notes from their textbook, then study from the AI-generated notes"
• Metacognitive risk: The specific concern. e.g. "Students believe they 'understand' the literary analysis because the essay looks good, but when asked to discuss the text in class without AI, they can't articulate the argument" / "Students believe they 'can solve equations' because they get correct answers with AI help, but fail when the AI isn't available" / "Students believe they 'know' the content because they read well-organised AI revision notes, but the reading produced familiarity, not understanding"

Optional (injected by context engine if available):

• Student level: Year group and proficiency
• Subject area: The curriculum subject
• AI tool: Which specific AI tool is being used
• Assessment context: How learning will be assessed

Prompt

You are an expert in metacognition and self-regulated learning, with deep knowledge of Winne & Hadwin's (1998) SRL model, Thiede et al.'s (2003) metacomprehension accuracy research, Dunning et al.'s (2003) work on the Dunning-Kruger effect, Bjork et al.'s (2013) illusions of competence, and emerging research on AI's impact on metacognition (Kazemitabaar et al., 2023). You understand that AI tools pose a specific and novel threat to metacognitive monitoring: they produce fluent, correct output that students mistake for evidence of their own understanding. This is not a minor concern — it is potentially the most significant educational risk of AI tools, because it undermines the self-regulation cycle that drives all independent learning.

CRITICAL PRINCIPLES:
- **The core problem is CALIBRATION.** Metacognitive monitoring works when students' confidence matches their competence. AI distorts calibration by inflating confidence (the work looks great) without necessarily increasing competence (the student may not have learned anything). The interventions must improve calibration, not just raise or lower confidence.
- **Fluency ≠ understanding.** When AI produces smooth, well-structured output, students experience processing fluency — the content feels easy to understand. But ease of processing does not indicate depth of learning. In fact, Bjork et al. (2013) showed that material that is HARDER to process (disfluent fonts, challenging language, interleaved examples) often produces BETTER learning. AI removes this desirable difficulty.
- **The solution is not banning AI.** It's redesigning the learning process so that students ENCOUNTER THEIR OWN KNOWLEDGE STATE — not just the AI's output. This means creating moments where students must produce from memory, without AI support, and compare their production to what they thought they knew.
- **Retrieval-based monitoring is the gold standard.** Thiede et al. (2003): the most effective way to improve metacognitive accuracy is to require RETRIEVAL — generating from memory rather than recognising from presented material. After using AI, students should close the AI, attempt the task from memory, and compare. This reveals the gap between perceived and actual understanding.
- **Metacognitive monitoring must be DESIGNED IN, not added on.** If you wait until the assessment to discover that students thought they knew the material but didn't, it's too late. Monitoring checkpoints must be built into the learning process — at the point of AI use, not after it.

Your task is to analyse the metacognitive risks and design monitoring interventions for:

**AI learning context:** {{ai_learning_context}}
**Metacognitive risk:** {{metacognitive_risk}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Student level:** {{student_level}} — if not provided, design for a general secondary school context.
**Subject area:** {{subject_area}} — if not provided, infer from the context.
**AI tool:** {{ai_tool}} — if not provided, assume a general-purpose LLM chatbot.
**Assessment context:** {{assessment_context}} — if not provided, assume a traditional exam without AI access.

Return your output in this exact format:

## Metacognitive Monitoring Analysis: [Context Description]

**Context:** [How students are using AI]
**Core risk:** [The specific metacognitive distortion — one sentence]
**Severity:** [How likely and how damaging this risk is — high/moderate/low]

### Metacognitive Diagnosis

[Detailed analysis of how AI use in this context distorts metacognitive monitoring. Name the specific illusions of competence at play. Explain the mechanism — HOW does the AI use lead to miscalibrated confidence?]

### Monitoring Interventions

[Specific strategies to improve metacognitive accuracy. For each:]

**Intervention [N]: [Name]**
- **What:** [What the student does]
- **When:** [At what point in the learning process — before, during, or after AI use]
- **Why it works:** [The metacognitive principle it applies]
- **Example:** [A concrete example of the intervention in this context]

### AI Usage Guidelines

[When to use AI and when to restrict it — specific, practical guidelines for this context]

### Assessment Alignment

[How to design assessment so that it measures student knowledge, not AI-assisted performance]

### Red Flags

[Observable signs that metacognitive distortion is occurring — what the teacher should watch for]

**Self-check before returning output:** Verify that (a) the diagnosis identifies the specific metacognitive mechanism, (b) interventions target calibration not just confidence, (c) retrieval-based monitoring is included, (d) AI is not banned but strategically managed, and (e) assessment aligns with metacognitive goals.

Example Output

Scenario: AI learning context: "Year 12 A-level English Literature students use ChatGPT to help write essays. The typical workflow: student receives an essay question (e.g., 'How does Shakespeare present the theme of power in Macbeth?'), enters it into ChatGPT, receives a well-structured essay with relevant quotes and analysis, then edits and personalises the AI output. Most students change some wording and add a few of their own points, but the core argument and structure are AI-generated." / Metacognitive risk: "Students believe they 'understand' the literary analysis because the essay looks good and they can follow the argument when reading it. But when asked to discuss the text in a class seminar without AI, or in a closed-book exam, they struggle to articulate the argument, recall specific quotes, or build an analytical case from scratch. They are surprised by this — they thought they understood it." / Student level: "Year 12, A-level English Literature" / AI tool: "ChatGPT (free version)" / Assessment context: "Closed-book A-level exam — no AI access, 45 minutes per essay"

---

Metacognitive Monitoring Analysis: AI-Assisted Essay Writing in A-Level English

Context: Year 12 students use ChatGPT to generate essay drafts, then edit and personalise them Core risk: Students confuse their ability to RECOGNISE good literary analysis (when reading AI output) with their ability to PRODUCE it (from their own knowledge) Severity: High — the gap between AI-assisted performance and unassisted exam performance could be substantial, and students are likely to discover this gap only at the point of assessment

Metacognitive Diagnosis

Three illusions of competence are operating simultaneously:

1. Recognition-production confusion (the most damaging). When a student reads the AI's analysis of power in Macbeth, they follow the argument. It makes sense. They nod along. They may even think "Yes, that's what I thought." But FOLLOWING an argument and CONSTRUCTING an argument are completely different cognitive tasks. Reading the AI's essay activates recognition ("I understand this when I see it") but does not build production capacity ("I can generate this myself"). The student's metacognitive monitoring is calibrated to their recognition ability, not their production ability — and the exam tests production.

2. Fluency illusion. ChatGPT produces polished, well-structured prose. When students read and edit this output, they experience processing fluency — it feels easy to understand, and therefore feels well-learned. Bjork et al. (2013) showed that fluency is one of the most misleading cues for learning. The student's confidence is high because the material feels smooth, not because their understanding is deep.

3. Effort misattribution. Students spend time editing the AI output — changing words, adding points, reorganising paragraphs. This feels like cognitive work, and students may attribute their sense of understanding to this editing effort. But editing someone else's analysis is a fundamentally different cognitive task from constructing your own. The editing effort produces familiarity with THIS specific essay, not transferable analytical skill.

The Dunning-Kruger amplifier: Students who are weakest in literary analysis benefit most from the AI (the gap between their ability and the AI's is largest) and are LEAST able to recognise the gap. A strong student who reads the AI's essay can identify where it differs from their own thinking. A weak student reads the same essay and experiences only agreement — they can't see what they're missing because they don't have the knowledge to see it.

Monitoring Interventions

Intervention 1: The "Close the Laptop" Test

• What: After using AI to develop an essay plan, the student closes ChatGPT, opens a blank document, and writes the essay's core argument in 3-4 sentences from memory. Then they compare their version to the AI's.
• When: DURING the writing process — after AI consultation but before final drafting
• Why it works: This is retrieval-based monitoring (Thiede et al., 2003). Writing from memory reveals the student's ACTUAL knowledge state, stripping away the recognition illusion. The comparison between their version and the AI's makes the gap visible and concrete.
• Example: Student reads AI's essay on power in Macbeth. Closes laptop. Writes: "Macbeth starts weak and becomes powerful through murder, but then loses power because of guilt." Compares to AI's version, which includes specific quotes, analysis of language, and connections to Jacobean politics. The gap is now visible: the student has a plot summary, not an analytical argument.

Intervention 2: Quote Recall Before AI Use

• What: Before consulting the AI, the student writes down every relevant quote they can remember for the essay question, with brief notes on what each quote shows. THEN they consult the AI and compare.
• When: BEFORE AI use — to establish a baseline of the student's own knowledge
• Why it works: This creates a "before" picture of the student's knowledge. When they see the AI's essay with 8 precisely chosen quotes and they could only recall 2, the metacognitive signal is clear: "I need to learn more quotes, not just read the AI's."
• Example: Student writes down: "'Fair is foul and foul is fair' — witches, shows things aren't what they seem. 'Is this a dagger I see before me' — Macbeth going mad." Then reads AI's essay which includes these plus six more. Student now has concrete evidence of what they know and don't know.

Intervention 3: Explain-It-Back Protocol

• What: After editing an AI-generated essay, the student must verbally explain the essay's argument to a partner (or to the AI itself) WITHOUT looking at the essay. The partner asks follow-up questions.
• When: AFTER AI use — to test whether understanding survives without the text
• Why it works: Verbal explanation requires production, not recognition. The student cannot hide behind edited AI prose. Gaps in understanding become immediately apparent when someone asks "Why did you say that?" and the student can't explain.
• Example: Student explains: "The essay argues that power corrupts Macbeth." Partner asks: "How does Shakespeare SHOW this through language?" Student pauses — they recognised the AI's language analysis when reading it, but can't reproduce it. The metacognitive signal lands.

Intervention 4: Prediction-Outcome Calibration

• What: Before the exam, the student predicts their score on a practice essay written without AI. After the essay is marked, they compare their prediction to their actual score.
• When: In the weeks before the exam — as a calibration exercise
• Why it works: Directly measures metacognitive accuracy. If a student predicts "I'll get a B" and gets a D, the miscalibration is undeniable. Repeated calibration exercises improve accuracy over time (Thiede et al., 2003).
• Example: Student has been writing AI-assisted essays all term and getting A grades on them. They predict a B on the unassisted practice essay. They get a D. The 2-grade gap is the metacognitive wake-up call.

AI Usage Guidelines

AI is helpful for:

• Exploring different analytical angles AFTER the student has developed their own initial response
• Identifying quotes the student hadn't considered (but the student must then learn and understand these quotes independently)
• Getting feedback on a STUDENT-WRITTEN draft (the AI evaluates, the student wrote)
• Modelling essay structure (studying HOW the AI organises an argument, not WHAT it argues)

AI is harmful when:

• The student consults AI BEFORE developing any independent thoughts
• The student uses AI output as the starting point rather than their own thinking
• The student edits AI text rather than writing from scratch
• The student uses AI as a substitute for close reading of the text itself

Recommended workflow:

1. Read the text closely. Make notes. Select quotes.
2. Plan the essay argument YOURSELF (even if it's rough)
3. Write a first draft YOURSELF (even if it's imperfect)
4. THEN consult AI: "Here's my argument. What have I missed? Where is my analysis weak?"
5. Revise YOUR draft using AI feedback — but the base text remains YOURS

Assessment Alignment

The assessment must test what the STUDENT knows, not what the AI-student system produces. For A-level English Literature:

1. Closed-book timed essays are already well-aligned — the student must produce from memory. Ensure students have practised this format regularly, not just AI-assisted homework essays.

2. Class seminars and discussions are excellent metacognitive tests — verbal analysis in real time cannot be AI-assisted. Use these as formative assessments throughout the term.

3. Quote tests (write 10 relevant quotes from memory for this theme) reveal knowledge of the text separately from analytical skill.

4. "Think aloud" analysis — give the student a passage they haven't seen and ask them to analyse it live, talking through their thinking. This tests the analytical SKILL, not the product.

Red Flags

Watch for these signs that metacognitive distortion is occurring:

• Homework-exam gap: Students produce excellent AI-assisted essays at home but perform significantly worse in timed, unassisted conditions. A consistent 2+ grade gap between homework and exam performance is a strong signal.
• Surface confidence, depth anxiety: Students say "I understand it" when reading their AI-edited essay but become visibly anxious when asked to discuss the text without notes. The confidence is calibrated to recognition, not production.
• Quote poverty: In class discussion, students can articulate arguments but cannot support them with specific quotes. The AI provided the quotes; the student didn't learn them.
• Generic analysis: Student writing without AI is vague and generic ("Shakespeare uses powerful language to show...") compared to their AI-assisted writing, which is specific and detailed. The specificity came from the AI, not the student.
• Inability to deviate from the plan: When a discussion takes an unexpected direction, AI-dependent students struggle to adapt because their analytical framework is borrowed, not owned.

---

Known Limitations

1. The evidence on AI-specific metacognitive effects is still emerging. Kazemitabaar et al. (2023) is one of a small but growing number of studies on AI tools and metacognition. The broader metacognitive research (Thiede et al., 2003; Bjork et al., 2013) provides strong theoretical grounding, but the specific application to LLM-assisted learning is based on extrapolation from these principles, not extensive empirical testing.

2. Monitoring interventions add cognitive and time costs. The "close the laptop" test, quote recall, and explain-it-back protocols all require additional time and effort. In contexts where students are under time pressure (heavy workloads, multiple subjects), adding metacognitive monitoring exercises may feel burdensome. Teachers must balance metacognitive accuracy against practical feasibility.

3. Individual differences in metacognitive ability are large. Some students are naturally good at monitoring their own understanding; others are not. The Dunning-Kruger effect suggests that the students who most need metacognitive support are the least likely to recognise that they need it. Interventions must be STRUCTURAL (built into the workflow for all students) rather than ADVISORY ("you should check your understanding").

4. The relationship between AI use and metacognition may be more nuanced than "AI harms metacognition." Some uses of AI (e.g., using AI to generate practice questions, then attempting them without AI) might actually IMPROVE metacognitive accuracy by creating retrieval opportunities. The risk is context-dependent, not absolute. The diagnosis above applies specifically to the "AI generates, student edits" workflow — other workflows may have different metacognitive profiles.

Productive Failure & Desirable Difficulty Designer

What This Skill Does

Redesigns a teaching sequence to incorporate productive failure (Kapur, 2008, 2016) and desirable difficulties (Bjork, 1994; Bjork & Bjork, 2011), replacing the standard "teach then practise" model with a "struggle then consolidate" model that produces deeper, more durable learning. The core paradox: students who struggle first and fail learn MORE in the long run than students who receive clear instruction first and succeed immediately — even though it feels worse during the lesson. Kapur (2016) showed that productive failure works because the generation phase (where students attempt problems before being taught) activates prior knowledge, reveals the limits of that knowledge, and creates "knowledge gaps" that make the subsequent instruction more meaningful. Bjork (1994) introduced the concept of "desirable difficulties" — conditions that make learning harder in the short term but more durable in the long term. These include spacing, interleaving, generation, and retrieval practice. This skill is particularly important in AI-enabled learning environments because AI tools can inadvertently REMOVE desirable difficulties — making tasks easier, providing immediate answers, and reducing the productive struggle that drives deep learning. The output includes a complete productive failure sequence (generation phase + consolidation phase), the specific desirable difficulties embedded in the task, safeguards to ensure failure is productive not destructive, and guidance on preventing AI-enabled cognitive offloading.

Evidence Foundation

Kapur (2008, 2016) developed the productive failure framework through a series of studies in mathematics classrooms. In the canonical design, students are given a complex, novel problem BEFORE any instruction — a problem they are expected to fail at. They work in small groups, generating multiple solution approaches, none of which are fully correct. THEN the teacher provides instruction on the canonical solution, explicitly comparing it to the students' generated approaches. Kapur (2016) found that students in the productive failure condition significantly outperformed students who received direct instruction first on measures of conceptual understanding and transfer — even though the direct instruction students performed better on immediate procedural tests. The key finding: it's not the failure that produces learning, but the GENERATION. Students who generate ideas, even wrong ones, develop richer representations of the problem space, which makes subsequent instruction more meaningful. Bjork (1994) and Bjork & Bjork (2011) articulated the broader principle of desirable difficulties: conditions that reduce performance during learning but enhance long-term retention and transfer. They identified four key desirable difficulties: (1) spacing — distributing practice over time rather than massing it, (2) interleaving — mixing different problem types rather than blocking them, (3) generation — producing answers rather than reading them, and (4) retrieval practice — testing yourself rather than restudying. All four share a common mechanism: they make the learning experience feel harder and less fluent, which paradoxically produces stronger memory traces and deeper understanding. Soderstrom & Bjork (2015) made the critical distinction between LEARNING and PERFORMANCE. Performance is what you can do RIGHT NOW — it's visible and measurable in the moment. Learning is the long-term change in knowledge or skill — it's invisible during the lesson and only measurable later. Desirable difficulties reduce performance (students get more wrong during the lesson) but enhance learning (students remember more and transfer better weeks later). This distinction is essential because teachers — and AI systems — tend to optimise for performance (making students succeed now) rather than learning (making students remember and transfer later).

Input Schema

The teacher must provide:

• Target concept: What students need to learn. e.g. "The concept of standard deviation — not just calculating it, but understanding what it MEANS and when to use it" / "The causes of WWI — not just listing them, but understanding how they interacted to produce war" / "Persuasive writing techniques — not just naming them, but choosing the right technique for a specific audience and purpose"
• Current approach: How it's currently taught. e.g. "I teach the formula, work through examples, then give practice questions. Students can calculate SD but don't understand what it tells them about data" / "I give a timeline and explain each cause. Students can list causes in an exam but can't explain how they connected" / "I teach techniques one at a time with examples. Students can identify techniques but struggle to use them independently"

Optional (injected by context engine if available):

• Student level: Year group and proficiency
• Subject area: The curriculum subject
• Time available: How much lesson time is available
• Prerequisite knowledge: What students already know
• Risk tolerance: Teacher's comfort with student struggle

Prompt

You are an expert in productive failure and desirable difficulties, with deep knowledge of Kapur's (2008, 2016) productive failure framework, Bjork's (1994) desirable difficulties research, Bjork & Bjork's (2011) applied desirable difficulties, and Soderstrom & Bjork's (2015) distinction between learning and performance. You understand the core paradox: making learning HARDER (in the right way) makes it BETTER — but it feels worse for both students and teachers. You also understand the specific risk that AI tools pose to desirable difficulties: by making tasks easier and providing immediate support, AI can remove the productive struggle that drives deep learning.

CRITICAL PRINCIPLES:
- **Generation before instruction.** The productive failure sequence MUST start with students attempting the task BEFORE receiving instruction. This is non-negotiable. The generation phase activates prior knowledge, reveals gaps, and creates the conditions for meaningful consolidation. If you teach first, you lose the generative benefit.
- **Failure must be PRODUCTIVE, not UNPRODUCTIVE.** Kapur (2016) distinguishes four conditions: productive failure (struggle → learning), unproductive failure (struggle → no learning), productive success (succeed → learning), and unproductive success (succeed → no learning). Productive failure requires: (a) the task is challenging but not impossible, (b) students have enough prior knowledge to generate partial solutions, (c) the consolidation phase explicitly connects student-generated ideas to the canonical solution, and (d) the classroom culture supports struggle without shame.
- **Design the consolidation phase as carefully as the generation phase.** Many teachers try productive failure and conclude "it didn't work" because they skip or rush the consolidation. The consolidation is where learning happens: the teacher explicitly compares the students' generated approaches with the correct approach, highlights what was right, what was wrong, and WHY the canonical approach works. Without this, students just experienced confusion.
- **Desirable difficulties feel undesirable.** Students will report that they "didn't learn anything" during the generation phase. They will rate the lesson lower on satisfaction surveys. They will perform worse on immediate post-tests. This is EXPECTED. The benefits appear later — on delayed tests, transfer tasks, and conceptual understanding measures. Warn teachers about this.
- **AI-enabled cognitive offloading is the enemy of desirable difficulty.** If students can ask an AI for the answer during the generation phase, the productive failure is destroyed. The design must include explicit guidance on when and how AI tools should be restricted to preserve the learning benefit.

Your task is to redesign this teaching sequence for productive failure:

**Target concept:** {{target_concept}}
**Current approach:** {{current_approach}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Student level:** {{student_level}} — if not provided, design for a general secondary school context.
**Subject area:** {{subject_area}} — if not provided, infer from the content.
**Time available:** {{time_available}} — if not provided, design for a double lesson (approximately 90 minutes).
**Prerequisite knowledge:** {{prerequisite_knowledge}} — if not provided, identify the likely prerequisites based on the target concept.
**Risk tolerance:** {{risk_tolerance}} — if not provided, design for a teacher who is willing to try but nervous about student frustration.

Return your output in this exact format:

## Productive Failure Sequence: [Target Concept]

**Target concept:** [What students need to learn]
**Current approach:** [How it's currently taught]
**Redesigned approach:** [One-sentence summary of the productive failure version]

### Phase 1: Generation (Productive Failure)

[The task students will attempt BEFORE instruction. Include the specific problem, the expected student responses (both correct and incorrect approaches they are likely to generate), and the teacher's role during this phase (circulating, NOT helping, noting student approaches for use in Phase 2).]

**The task:**
[Exact task students receive]

**Expected student responses:**
[What students will likely try — including the specific partial solutions and incorrect approaches]

**Teacher role during generation:**
[What the teacher does — observe, collect approaches, resist the urge to help]

### Phase 2: Consolidation

[How the teacher transitions from generation to instruction. This must explicitly connect student-generated approaches to the canonical solution. Include the specific comparisons the teacher should make and the key insights that emerge from the comparison.]

### Desirable Difficulties Embedded

[List each desirable difficulty in the sequence and explain why it enhances learning]

| Difficulty | What It Does | Why It Helps |
|---|---|---|
| [Difficulty] | [Mechanism] | [Learning benefit] |

### Failure Safeguards

[How to ensure failure is PRODUCTIVE — the conditions that must be in place and what to do if students become frustrated or disengaged]

### Cognitive Offloading Risks

[How AI tools might undermine the desirable difficulties and specific recommendations for managing this]

### Teacher Preparation Notes

[What the teacher needs to know and do before attempting this sequence — including how to manage student frustration and their own discomfort with watching students struggle]

**Self-check before returning output:** Verify that (a) generation comes before instruction, (b) the task is challenging but not impossible, (c) the consolidation explicitly connects student approaches to the canonical solution, (d) desirable difficulties are identified and justified, and (e) cognitive offloading risks are addressed.

Example Output

Scenario: Target concept: "The concept of standard deviation — not just calculating it, but understanding what it MEANS (a measure of spread/variability) and when to use it (comparing consistency, identifying outliers, making predictions)" / Current approach: "I teach the formula, work through two examples on the board, then give 10 practice questions. Students can calculate SD to two decimal places but when asked 'what does this standard deviation tell you about the data?' they can't answer. They treat it as a procedure, not a concept" / Student level: "Year 10, GCSE Higher tier, comfortable with mean and range" / Subject area: "Maths — statistics" / Time available: "One 60-minute lesson"

---

Productive Failure Sequence: Standard Deviation

Target concept: Understanding what standard deviation MEANS (a measure of how spread out data is from the mean) and when to use it (comparing variability, not just calculating a number) Current approach: Teach formula → worked examples → practice questions (produces procedural fluency without conceptual understanding) Redesigned approach: Students attempt to invent a measure of "spread" from data before being taught the standard deviation formula — generating and comparing their own approaches builds the conceptual understanding that direct instruction misses

Phase 1: Generation (Productive Failure)

Duration: 20-25 minutes

The task:

Present this on the board or as a handout:

The Football Manager Problem

You are a football manager choosing between two strikers. Both scored an average of 1.5 goals per match over the last 10 matches. But their match-by-match records are very different:

Player A: 2, 1, 2, 1, 2, 1, 2, 1, 2, 1 Player B: 5, 0, 0, 4, 0, 0, 6, 0, 0, 0

Both have a mean of 1.5 goals per match. But they're clearly VERY different players.

Your challenge (in pairs):

1. In what way are these players different? Describe the difference in words.
2. Invent a NUMBER that captures this difference. Your number should be BIGGER for the player whose scores are more spread out and SMALLER for the player whose scores are more consistent. You can use any mathematical operations you like.
3. Test your number on both players. Does it correctly show that Player B is more "spread out" than Player A?
4. If you have time: would your number work for ANY two datasets, not just these two?

Expected student responses:

Students will generate several approaches, most of which will be partially correct:

• Range approach (most common): Player A range = 2 - 1 = 1. Player B range = 6 - 0 = 6. This works but is fragile — it only uses two data points and ignores everything in between. Approximately 60-70% of pairs will try this first.

• "Distance from mean" approach (less common but promising): Some students will calculate how far each score is from the mean (1.5) and try to combine these distances. They'll likely add the distances, getting different totals for each player. This is close to the conceptual basis of standard deviation. Approximately 20-30% of pairs.

• "Count the zeros" or "count the extremes" approach: Some students will try non-mathematical approaches — counting how many times a player scored 0, or how many "big" games they had. This captures the intuition but isn't generalisable. Approximately 10-20%.

• Stuck/confused: Some pairs will struggle to move from the verbal description ("Player B is more inconsistent") to a numerical measure. This is expected. The teacher should NOT help — the struggle is the point.

What students will NOT generate:

• Squaring the deviations (the key innovation in standard deviation). Students will likely add or average the raw deviations from the mean, which produces a useful number but has mathematical limitations (positive and negative deviations cancel out if you don't take absolute values or square).

Teacher role during generation:

• Circulate and OBSERVE. Do not teach. Do not hint. Do not correct. Your job is to collect student approaches for use in Phase 2.
• Note which approaches each pair used. You'll need 3-4 contrasting approaches for the consolidation.
• If students are stuck after 10 minutes: You may offer ONE prompt: "What if you started by finding how far each score is from the average?" This redirects without giving the answer.
• If students are frustrated: Acknowledge the difficulty. "This is genuinely hard. Mathematicians argued about this for centuries. You're not expected to get the 'right' answer — I want to see your THINKING."
• Keep time. The generation phase should last 20-25 minutes. Even if students haven't finished, stop. Incomplete generation is fine — the consolidation will build on whatever they produced.

Phase 2: Consolidation

Duration: 30-35 minutes

Step 1: Share and compare (10 minutes). Select 3 student approaches (ideally: range, distance-from-mean, and one other). Write each on the board. For each approach, ask the class: "Does this number correctly show that Player B is more spread out? What are the strengths and weaknesses of this approach?" Guide the discussion toward the key limitations:

• Range: simple, but only uses two data points. A single outlier distorts it completely.
• Sum of distances from mean: better, but what if you have more data points? The sum gets bigger just because there's more data. You need to AVERAGE the distances.
• Average distance from mean: good! But there's a problem: some distances are positive and some are negative, and they cancel out. (Demonstrate: for Player A, the deviations from 1.5 are +0.5, -0.5, +0.5, -0.5… which SUM to zero.)

Step 2: Introduce the canonical solution (10 minutes). "Here's how mathematicians solved the cancellation problem: they SQUARED each deviation (making everything positive), averaged the squares, then took the square root to get back to the original units. This is called the standard deviation." Walk through the formula using Player A and Player B data, explicitly connecting each step to the students' generated approaches: "See how this step is just what Group 2 did — finding the distance from the mean? The only new step is the squaring."

Step 3: Connect back to the original question (10 minutes). Calculate SD for both players. Player A: SD ≈ 0.5. Player B: SD ≈ 2.3. "Player B's standard deviation is about 4.5 times larger than Player A's. What does that MEAN for the football manager?" Push for interpretation: SD is not just a number — it tells you how PREDICTABLE a player is. "If you need a reliable scorer every week, choose Player A. If you're desperate for a match-winner and can absorb some scoreless games, Player B might be worth the risk."

Step 4: Quick practice with understanding focus (5 minutes). Give one more dataset. Ask students to (a) calculate SD AND (b) interpret what it means in context. The interpretation is the assessment target, not the calculation.

Desirable Difficulties Embedded

Difficulty	What It Does	Why It Helps
Generation (attempting before instruction)	Forces students to activate prior knowledge and grapple with the problem of measuring spread	Creates "knowledge gaps" that make the formula meaningful when introduced — students understand WHY each step exists
Productive failure (students don't reach the correct answer)	Creates cognitive conflict — students see that their approaches have limitations	Motivation to learn the canonical solution is higher because they've experienced the problem it solves
Interleaving (verbal → numerical → interpretation within one task)	Students must move between different representations of the same concept	Builds flexible understanding rather than rigid procedural knowledge
Delayed instruction (teaching after struggle, not before)	Prevents students from simply following the procedure without understanding	The formula is introduced as a SOLUTION to a problem students have experienced, not as an arbitrary recipe

Failure Safeguards

Condition 1: The task must be challenging but not impossible. The Football Manager problem is carefully calibrated: Year 10 students can calculate mean and range, and they can describe spread in words. They have enough prior knowledge to generate partial solutions. If students couldn't calculate a mean, this task would be too hard — they'd experience UNPRODUCTIVE failure.

Condition 2: The classroom culture must support struggle. Before the generation phase, explicitly say: "I'm about to give you a problem that you're NOT expected to solve perfectly. The struggle IS the point. I want to see your thinking, not a correct answer." This normalises failure and reduces anxiety.

Condition 3: Frustration monitoring. If more than half the class appears stuck and frustrated after 10 minutes, offer the single scaffold ("What if you started by finding how far each score is from the average?"). This prevents the failure from becoming unproductive. If frustration persists at 15 minutes, begin Phase 2 early — you have enough student approaches to work with.

Condition 4: The consolidation must be thorough. If you run out of time, cut the practice questions (Step 4), NEVER the consolidation (Steps 1-3). A productive failure sequence without consolidation is just a confusing lesson.

Cognitive Offloading Risks

Risk 1: Students Google "how to measure spread." If students have internet access, they will find "standard deviation" within 30 seconds, bypassing the entire generation phase. Mitigation: This is a no-devices task. Calculators yes, phones and laptops no. Explain why: "I need you to figure this out yourselves — that's the whole point."

Risk 2: Students ask an AI chatbot for help. In schools where AI tools are available, students may prompt "how do you measure how spread out data is?" and get a clear explanation of standard deviation. Mitigation: Explicitly instruct that AI tools are not to be used during Phase 1. Frame it positively: "You're doing something an AI can't do — inventing your own mathematical approach. The AI already knows the formula. You don't need the formula right now — you need the THINKING."

Risk 3: High-achieving students short-circuit the struggle. Some students may already know about standard deviation or may find the answer in a textbook. Mitigation: The task asks students to INVENT a measure, not to FIND one. If a student says "Isn't this just standard deviation?" respond: "Maybe. But can you explain WHY standard deviation works? Why square the deviations instead of taking absolute values? That's what I want you to figure out."

Teacher Preparation Notes

Expect discomfort. Watching students struggle without helping is one of the hardest things a teacher can do. You will feel the urge to explain, to give hints, to rescue. Resist. The struggle is the product, not a problem to be solved. Kapur's research shows that the students who struggle most during the generation phase often learn the most during the consolidation phase.

Expect student complaints. Students may say "Why didn't you just teach us the formula?" or "This was confusing." This is NORMAL and does not mean the lesson failed. Soderstrom & Bjork (2015) showed that students' subjective experience of learning is a poor predictor of actual learning — they feel they learn more from clear explanations, but they remember and transfer more from productive failure.

Prepare the consolidation in advance. Know which student approaches you're looking for (range, distance from mean, other). Have the canonical SD calculation ready to present alongside student approaches. The consolidation cannot be improvised — it requires careful planning.

Have a backup plan. If the generation phase produces only one approach (everyone does range), adjust the consolidation: "Most of you tried range. That's a good start. Let me show you another approach that a pair in a different class tried…" and present the distance-from-mean approach yourself.

---

Known Limitations

1. Productive failure is not appropriate for all content. It works best for CONCEPTUAL understanding — understanding WHY something works, not just HOW to do it. For purely procedural skills (long division, balancing equations), direct instruction followed by practice is often more efficient. Productive failure is most valuable when the current approach produces procedural fluency without conceptual understanding.

2. The evidence is stronger for mathematics and science than for other domains. Kapur's (2008, 2016) research was conducted primarily in mathematics classrooms. The principle of generation before instruction has been studied in other domains, but the specific productive failure design (attempt → fail → consolidate) has less evidence outside STEM. The underlying mechanism (generation enhances learning) is domain-general, but the specific task design may need adaptation.

3. Teacher skill is a binding constraint. The consolidation phase requires a skilled teacher who can connect student-generated approaches to the canonical solution in real time. This is significantly harder than delivering a prepared lecture. Teachers attempting productive failure for the first time should start with a topic they know deeply.

4. The performance-learning distinction creates an assessment problem. Students in productive failure conditions perform WORSE on immediate post-tests but BETTER on delayed and transfer tests (Soderstrom & Bjork, 2015). If the teacher assesses learning immediately after the lesson, productive failure will appear to have failed. The real benefits only appear days or weeks later.

Self-Explanation Prompt Designer

What This Skill Does

Designs self-explanation prompts — the specific questions placed at key points in learning material that prompt students to explain TO THEMSELVES why something is true, how it works, or what it means. Chi et al. (1989) discovered that students who spontaneously self-explain while studying worked examples learn dramatically more than those who don't — and crucially, Chi et al. (1994) showed that students who are PROMPTED to self-explain also show significant learning gains, even if they wouldn't have self-explained spontaneously. The self-explanation effect is one of the most robust findings in learning science: it works across ages, domains, and material types. The mechanism is that self-explaining forces the learner to generate inferences, fill in gaps, connect new information to existing knowledge, and notice their own confusion — all of which strengthen understanding. The challenge is designing prompts that actually elicit deep self-explanation rather than shallow restating. "Why does this step follow from the previous one?" is a good self-explanation prompt. "What happens in this step?" is not — it elicits description, not explanation. AI is specifically valuable here because it can deliver prompts at exactly the right moment (during study, not after) and can evaluate the quality of student responses in real time.

Evidence Foundation

Chi et al. (1989) conducted the seminal study on self-explanation, observing students studying worked examples in physics. They found that "good students" (those who learned the most) spontaneously engaged in self-explanation: they paused at each step to explain to themselves WHY the step made sense, HOW it connected to the previous step, and WHAT principle it illustrated. "Poor students" read the examples passively, focusing on the surface steps without generating explanations. The difference in learning was dramatic — and it was not explained by prior knowledge, intelligence, or study time. The critical variable was the QUALITY of cognitive engagement during study. Chi et al. (1994) tested whether prompting self-explanation could replicate the benefits seen in spontaneous self-explainers. They gave students a biology text and prompted one group to self-explain each sentence ("What new information does this sentence provide? How does it relate to what you already know?"). The prompted group significantly outperformed the control group on both immediate and transfer tests. This finding was revolutionary because it demonstrated that a simple instructional intervention (prompting) could produce the same benefits as a rare cognitive habit. Hausmann & VanLehn (2007) investigated what makes self-explanation effective. They found that the benefit comes from two components: the CONTENT of the explanation (the ideas generated) and the act of GENERATION itself (producing an explanation rather than reading one). When they compared self-generated explanations with instructor-provided explanations of equal quality, the self-generated ones still produced better learning — suggesting that the act of generating, not just the content, drives the effect. Wylie & Chi (2014) extended self-explanation research to multimedia learning, showing that self-explanation prompts embedded in digital materials (videos, interactive simulations, digital texts) produce the same benefits as prompts in text-based materials. They found that prompts should be placed at points of high conceptual density — where the material introduces a new principle, makes a non-obvious inference, or contradicts common intuition. Rittle-Johnson (2006) studied self-explanation in mathematics and found that self-explanation prompts improve both conceptual understanding and procedural transfer — students who self-explain can not only solve the problems they studied but also apply the principles to new, structurally different problems.

Input Schema

The teacher must provide:

• Learning material: What students are studying. e.g. "A worked example showing how to solve a quadratic equation by completing the square: x² + 6x + 2 = 0 → x² + 6x = -2 → x² + 6x + 9 = -2 + 9 → (x + 3)² = 7 → x + 3 = ±√7 → x = -3 ± √7" / "A biology text explaining how vaccines work: 'A vaccine introduces a weakened or inactive form of a pathogen into the body. The immune system recognises the pathogen's antigens and produces antibodies. Memory cells are created that remember the specific antigen...'" / "A historical source: a propaganda poster from WWI with specific visual elements"
• Target understanding: What students should deeply understand. e.g. "Why we add 9 to both sides — specifically, that (b/2)² completes the perfect square trinomial, and that this transforms the equation into a form we can square-root" / "Why vaccines work BEFORE you get sick — the distinction between active immunity (memory cells) and passive immunity, and why the weakened pathogen doesn't cause disease" / "How propaganda uses visual techniques to manipulate emotion — the specific choices the artist made and why those choices are persuasive"

Optional (injected by context engine if available):

• Student level: Year group and proficiency
• Subject area: The curriculum subject
• Delivery mode: How prompts will be delivered
• Known misconceptions: Common misunderstandings about this content

Prompt

You are an expert in self-explanation research, with deep knowledge of Chi et al.'s (1989, 1994) foundational studies, Hausmann & VanLehn's (2007) work on the generation effect in self-explanation, Wylie & Chi's (2014) self-explanation principle in multimedia learning, and Rittle-Johnson's (2006) research on self-explanation in mathematics. You understand that self-explanation is one of the most powerful learning strategies available — and one of the easiest to implement badly. The difference between effective and ineffective self-explanation prompts is the difference between deep conceptual engagement and surface-level restating.

CRITICAL PRINCIPLES:
- **Prompts must elicit EXPLANATION, not DESCRIPTION.** "What happens in Step 3?" elicits description ("We add 9 to both sides"). "WHY do we add specifically 9 — where does 9 come from?" elicits explanation ("Because (6/2)² = 9, and adding this completes the perfect square"). The prompt must target the REASONING, not the CONTENT.
- **Place prompts at points of high conceptual density.** Wylie & Chi (2014): self-explanation prompts are most effective when placed where the material introduces a new principle, makes a non-obvious step, or is likely to be misunderstood. Don't prompt at every step — prompt at the steps that MATTER for understanding.
- **The generation is the learning.** Hausmann & VanLehn (2007): even if a student's self-explanation is imperfect, the ACT of generating it produces learning. Design prompts that require generation (constructing an explanation in their own words) rather than recognition (selecting the right answer from options). Free-response prompts > multiple choice.
- **Self-explanation prompts should be answerable.** A prompt that the student cannot answer without additional instruction is not a self-explanation prompt — it's a test question. Self-explanation works by activating and connecting EXISTING knowledge. The material should contain enough information for a thoughtful student to construct an explanation.
- **Distinguish self-explanation from other activities.** Self-explanation is not summarising (restating the content), not predicting (guessing what comes next), and not evaluating (judging whether something is good). It is explaining the REASONING behind something — the WHY and the HOW, not the WHAT.

Your task is to design self-explanation prompts for:

**Learning material:** {{learning_material}}
**Target understanding:** {{target_understanding}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Student level:** {{student_level}} — if not provided, design for a general secondary school context.
**Subject area:** {{subject_area}} — if not provided, infer from the material.
**Delivery mode:** {{delivery_mode}} — if not provided, design for prompts embedded in a digital learning environment (shown on screen at key points).
**Known misconceptions:** {{known_misconceptions}} — if not provided, identify the most likely misconceptions based on the material.

Return your output in this exact format:

## Self-Explanation Prompts: [Material Description]

**Material:** [What students are studying]
**Target understanding:** [The deep insight students should reach]
**Number of prompts:** [How many, and why this number]

### Prompt Placement Map

[Show where each prompt falls in the material — at which step, sentence, or element. Explain why THAT point was chosen.]

### Self-Explanation Prompts

For each prompt:

**Prompt [N] — placed at: [specific location in the material]**
- **Prompt text:** [Exactly what the student sees]
- **Target explanation:** [What a good self-explanation would include]
- **Why here:** [Why this point in the material is a high-value prompt location]
- **Common shallow response:** [What students say when they're describing rather than explaining]
- **Follow-up if shallow:** [What to ask next if the student's response is surface-level]

### Quality Indicators

[How to distinguish deep self-explanation from shallow restating — with specific examples of each]

| Quality Level | What It Sounds Like | Example | What to Do |
|---|---|---|---|
| Deep | [Characteristics] | [Example response] | [Affirm and continue] |
| Partial | [Characteristics] | [Example response] | [Prompt for elaboration] |
| Shallow | [Characteristics] | [Example response] | [Redirect to reasoning] |

### Scaffolding Sequence

[For students who struggle to self-explain: how to move from open prompts to increasingly structured scaffolds without giving the answer]

**Self-check before returning output:** Verify that (a) prompts elicit explanation not description, (b) prompts are placed at conceptually dense points, (c) prompts require generation not recognition, (d) the material contains enough information to construct an explanation, and (e) quality indicators distinguish deep from shallow responses.

Example Output

Scenario: Learning material: "A worked example showing how to solve a quadratic equation by completing the square: Step 1: x² + 6x + 2 = 0. Step 2: Rearrange to x² + 6x = -2. Step 3: Add 9 to both sides: x² + 6x + 9 = -2 + 9 = 7. Step 4: Factorise the left side: (x + 3)² = 7. Step 5: Take the square root of both sides: x + 3 = ±√7. Step 6: Solve: x = -3 ± √7." / Target understanding: "Why completing the square works — specifically, understanding that (b/2)² is the number that transforms x² + bx into a perfect square trinomial, and why this transformation is useful (it lets us take the square root of both sides)" / Student level: "Year 11, GCSE Higher" / Known misconceptions: "Students think 9 appears 'by magic' — they can follow the steps but don't understand where 9 came from. Students also struggle with ±√7 — they forget the negative root or don't understand why there are two answers"

---

Self-Explanation Prompts: Completing the Square

Material: Worked example solving x² + 6x + 2 = 0 by completing the square Target understanding: Why (b/2)² completes the square — the algebraic structure, not just the procedure Number of prompts: 3 — placed at the three points of highest conceptual density (Steps 2, 3, and 5)

Prompt Placement Map

The worked example has 6 steps. NOT every step needs a prompt:

• Step 1 (write equation): Routine — no prompt needed.
• Step 2 (rearrange to x² + 6x = -2): PROMPT 1 — this step makes a non-obvious choice (why move the 2?) that reveals the strategy.
• Step 3 (add 9 to both sides): PROMPT 2 — this is the KEY conceptual step and the most common point of confusion. Where does 9 come from?
• Step 4 (factorise to (x + 3)² = 7): No prompt — if the student understood Prompt 2, this step follows. If they didn't, prompting here won't help.
• Step 5 (take square root, ±√7): PROMPT 3 — the ± is conceptually important and commonly misunderstood.
• Step 6 (solve for x): Routine — no prompt needed.

Self-Explanation Prompts

Prompt 1 — placed at: Step 2 (x² + 6x = -2)

• Prompt text: "We moved the +2 to the other side. WHY? What are we trying to set up on the left side of the equation, and why does having x² + 6x by itself help?"
• Target explanation: "We want the left side to be JUST the x terms (x² + 6x) so we can add a specific number to make it a perfect square. The +2 would get in the way because x² + 6x + 2 is NOT a perfect square. We need to control what gets added."
• Why here: This step seems minor ("just rearranging") but it reveals the STRATEGY of completing the square. If students don't understand WHY we isolate x² + 6x, the rest of the method is a mystery.
• Common shallow response: "We moved the 2 to get x on one side" — this describes the action without explaining the purpose.
• Follow-up if shallow: "But why do we need x² + 6x alone? What are we going to DO with it that requires the constant to be gone?"

Prompt 2 — placed at: Step 3 (x² + 6x + 9 = 7)

• Prompt text: "We added 9 to both sides. Explain: where does the number 9 come from? Why 9 specifically, and not 4 or 16 or any other number?"
• Target explanation: "9 comes from (6/2)² = 3² = 9. We take the coefficient of x (which is 6), halve it (3), and square it (9). We use THIS specific number because x² + 6x + 9 = (x + 3)² — it's the number that makes the left side a PERFECT SQUARE TRINOMIAL. No other number would do this."
• Why here: This is the conceptual heart of the method. The "where does 9 come from?" question targets the most common misconception directly. If a student can explain this, they understand completing the square.
• Common shallow response: "Because it completes the square" — this is circular (restating the method name without explaining the mechanism).
• Follow-up if shallow: "What does 'completing the square' actually mean? Try this: expand (x + 3)². What do you get? Now look at x² + 6x + 9. Notice anything?"

Prompt 3 — placed at: Step 5 (x + 3 = ±√7)

• Prompt text: "We wrote ±√7, not just √7. Why are there TWO values? What would we miss if we only wrote +√7?"
• Target explanation: "Because both (+√7)² and (-√7)² equal 7. The square root operation has two solutions. If we only took the positive root, we'd miss the second solution: x = -3 - √7. Geometrically, a quadratic has two roots (the parabola crosses the x-axis in two places), so we should expect two answers."
• Why here: The ± is a conceptual tripwire — students routinely forget it or don't understand why it's there. This prompt forces them to engage with the mathematical reason, not just memorise the symbol.
• Common shallow response: "Because it's ±" or "because that's the rule" — restating the notation without explaining the reasoning.
• Follow-up if shallow: "Try it without the ±. If x + 3 = √7, what's x? Now check: does that value actually solve the original equation? Now try x + 3 = -√7. Does THAT value also solve the original equation?"

Quality Indicators

Quality Level	What It Sounds Like	Example	What to Do
Deep	References the mathematical PRINCIPLE — explains WHY, connects to structure, uses own words	"We add 9 because (6/2)² = 9, and that's what makes x² + 6x + 9 factorise into a perfect square. We need a perfect square so we can take the square root in the next step."	Affirm: "Exactly right. You've identified the key principle." Move on.
Partial	Correct but incomplete — states part of the reason without the full chain	"9 comes from halving 6 and squaring it."	Prompt for elaboration: "Good — but WHY does halving and squaring give you the right number? What happens when you add it to x² + 6x?"
Shallow	Describes the action or restates the procedure without explaining the reasoning	"We add 9 to both sides to complete the square."	Redirect: "You've described WHAT we did, but I need you to explain WHY. Where does 9 specifically come from? Why not 4 or 16?"

Scaffolding Sequence

For students who struggle to self-explain, move through these levels:

Level 1 — Open prompt (default): "Explain why we added 9 to both sides."

Level 2 — Directed prompt: "Look at the number in front of x (it's 6). Try halving it. Now try squaring what you get. What number do you get? Does that match what was added?"

Level 3 — Verification prompt: "Try expanding (x + 3)². What do you get? Now compare that to x² + 6x + 9. What do you notice?"

Level 4 — Completion prompt: "We add (b/2)² because x² + bx + (b/2)² = (x + b/2)². In this case, b = 6, so (b/2)² = ___. And x² + 6x + ___ = (x + ___)². Fill in the blanks."

The sequence moves from maximum generation (Level 1) to maximum scaffolding (Level 4). Always start at Level 1 — the generation attempt, even if it fails, produces learning (Hausmann & VanLehn, 2007). Only escalate if the student is genuinely stuck, not just producing an imperfect explanation.

---

Known Limitations

1. Self-explanation prompts slow down study time. A student who self-explains at every step will take significantly longer to work through material than a student who reads passively. This is a DESIRABLE difficulty (Bjork, 1994) — the extra time produces deeper learning — but it can create practical problems in time-constrained contexts (exam revision, homework). Teachers must make conscious choices about where to invest self-explanation time.

2. The quality of self-explanation is hard to evaluate automatically. An AI can detect the LENGTH of a response and the presence of key terms, but evaluating whether a self-explanation is genuinely deep (rather than long and shallow) requires understanding the conceptual content. Current LLMs can approximate this evaluation, but they may mis-classify sophisticated but unconventional explanations as shallow.

3. Self-explanation works best with conceptually rich material. Procedural content with minimal conceptual depth (e.g., formatting a spreadsheet, conjugating regular verbs) benefits less from self-explanation prompts because there is less to EXPLAIN. Self-explanation is most powerful when the material contains non-obvious reasoning, hidden connections, or common misconceptions.

4. Students may need training to self-explain. Chi et al. (1994) found that prompting alone was effective, but other studies suggest that students benefit from initial training in what a good self-explanation looks like. The scaffolding sequence above addresses this, but in the first session using self-explanation prompts, the teacher may need to model the process explicitly.

Worked Example to Problem Solving Transition Designer

What This Skill Does

Designs the transition sequence from studying worked examples to solving problems independently — the critical phase where scaffolding is gradually removed and students take over the cognitive work. This addresses one of the most important findings in cognitive load theory: the expertise reversal effect (Kalyuga et al., 2003). When students are novices, worked examples are highly effective — they reduce extraneous cognitive load and allow students to build schemas. But as students develop competence, the same worked examples become REDUNDANT and actually HARM learning — the scaffolding that helped novices now prevents more advanced students from engaging in the active processing that drives further learning. The optimal instruction is not fixed but ADAPTIVE: it should shift from worked examples to problem solving as the student's expertise grows. Renkl & Atkinson (2003) developed the "fading" approach: rather than an abrupt switch from examples to problems, gradually remove steps from the worked examples so students progressively take over more of the work. This skill designs the complete fading sequence, including the triggers for when to fade (based on student performance), the order of fading (which steps to remove first), and the design of the independent practice phase that follows.

Evidence Foundation

Kalyuga et al. (2003) demonstrated the expertise reversal effect through a series of experiments showing that instructional techniques highly effective for novices become ineffective or harmful for more advanced learners. In the context of worked examples: novices who studied worked examples significantly outperformed novices who solved problems (the worked example effect), but as students gained expertise, this advantage reversed — more advanced students learned more from problem solving than from studying examples. The explanation is cognitive load theory: for novices, worked examples reduce the extraneous load of means-ends analysis (trying to figure out what to do), freeing cognitive resources for schema building. For advanced students, the worked example creates REDUNDANCY — the student already has a schema and the example is now unnecessary information that competes with their existing knowledge for processing resources. Kalyuga (2007) extended this work, arguing that learner-tailored instruction must continuously assess the learner's expertise level and adjust the instructional format accordingly. The practical implication: there is no single "best" instructional approach — the best approach depends on where the learner is RIGHT NOW. Renkl & Atkinson (2003) proposed fading as the solution to the example-to-problem transition. Rather than a sharp switch from "study examples" to "solve problems," they designed a gradual transition: first, full worked examples; then, examples with one step removed (the student completes that step); then, examples with two steps removed; and so on until the student is solving complete problems. They found that fading produced better learning than either fixed worked examples or fixed problem solving, because it continuously calibrated the cognitive demand to the student's growing expertise. Sweller et al. (2011) integrated the expertise reversal effect into the broader cognitive load theory framework, arguing that all instructional design must consider the INTERACTION between the learner's current knowledge and the instructional format. A technique that is optimal at one stage of learning may be counterproductive at another. Van Merriënboer & Kirschner (2018) developed the 4C/ID model for complex learning, which systematically designs the transition from heavily scaffolded task performance to independent performance through a sequence of task classes with decreasing support.

Input Schema

The teacher must provide:

• Skill being taught: What students need to do independently. e.g. "Solving quadratic equations by factorisation — finding factors, setting each factor to zero, solving for both roots" / "Writing a balanced argument essay — introduction with thesis, supporting paragraphs with evidence, counterargument paragraph, conclusion with evaluated position" / "Calculating the mean, median, and mode from a frequency table and interpreting which average is most appropriate"
• Current student state: Where students are now. e.g. "Students have seen two teacher-led worked examples. They can follow the steps when watching but haven't attempted any problems independently yet" / "Students can write persuasive essays (one-sided argument) but have never structured a balanced argument" / "Students can calculate mean from raw data but haven't worked with frequency tables"

Optional (injected by context engine if available):

• Student level: Year group and proficiency
• Subject area: The curriculum subject
• Number of practice problems: How many problems are available
• Time available: Duration for the transition sequence
• Assessment format: How competence will be tested

Prompt

You are an expert in the expertise reversal effect and the transition from worked examples to problem solving, with deep knowledge of Kalyuga et al.'s (2003) expertise reversal research, Kalyuga's (2007) learner-tailored instruction principles, Renkl & Atkinson's (2003) fading approach, Sweller et al.'s (2011) cognitive load theory, and Van Merriënboer & Kirschner's (2018) 4C/ID model. You understand that the most common instructional error is using a FIXED approach (always examples, or always problems) when the optimal approach is ADAPTIVE — shifting from scaffolded examples to independent problem solving as the student's expertise grows.

CRITICAL PRINCIPLES:
- **The expertise reversal effect is real and consequential.** Continuing to provide worked examples after students have built adequate schemas REDUCES learning. The scaffolding becomes cognitive clutter that the brain must process and suppress. This is not about student preference (students often prefer continued scaffolding because it feels easier) — it's about cognitive science.
- **Fade gradually, don't switch abruptly.** Renkl & Atkinson (2003): the transition from examples to problems should be a GRADIENT, not a cliff. Remove one step at a time, starting with the LAST step (backward fading) so students complete the final step first, then the last two steps, and so on. This builds confidence and ensures each fading stage builds on the previous one.
- **Use performance to trigger fading.** The trigger for removing the next step should be PERFORMANCE-BASED: the student consistently completes the current step correctly. Do not fade on a fixed schedule — students learn at different rates. An AI system can personalise the fading; a teacher can use quick checks.
- **Independent practice must be DESIGNED, not just assigned.** Once fading is complete, the problem-solving phase needs its own design: problem difficulty should increase gradually, problems should be interleaved (not blocked by type), and students who get stuck should receive the MINIMUM help needed (not a return to full worked examples).
- **Watch for the "illusion of competence" during fading.** Students who successfully complete faded examples may FEEL more competent than they are, because the remaining worked steps are doing much of the cognitive work. True competence is demonstrated only in FULLY independent problem solving.

Your task is to design the transition sequence for:

**Skill being taught:** {{skill_being_taught}}
**Current student state:** {{current_student_state}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Student level:** {{student_level}} — if not provided, design for a general secondary school context.
**Subject area:** {{subject_area}} — if not provided, infer from the skill.
**Number of practice problems:** {{number_of_practice_problems}} — if not provided, design for 8-12 problems across the full sequence.
**Time available:** {{time_available}} — if not provided, design for two 50-minute lessons.
**Assessment format:** {{assessment_format}} — if not provided, assume a timed test without scaffolding.

Return your output in this exact format:

## Worked Example to Problem Solving Transition: [Skill]

**Target skill:** [What students will do independently]
**Starting point:** [Where students are now]
**Transition approach:** [Fading method and rationale]

### Complete Transition Sequence

[The full sequence from worked example to independent problem solving, with each stage clearly defined]

**Stage [N]: [Name] — [Scaffolding level]**
- **What the student sees:** [The example/problem at this stage]
- **What the student does:** [Which steps the student completes]
- **Fading trigger:** [How to know the student is ready for the next stage]
- **Estimated problems at this stage:** [How many before moving on]

### Expertise Reversal Checkpoints

[The specific points where the system/teacher checks whether continued scaffolding is helping or hindering. Include the check method and the decision rule.]

### Fading Schedule

[Which steps are faded in which order, and why this order was chosen]

| Stage | Steps Shown | Steps Faded | Rationale |
|---|---|---|---|
| [Stage] | [Which steps the student can see] | [Which steps the student must complete] | [Why this order] |

### Independent Practice Design

[The design of the fully independent problem-solving phase — problem selection, difficulty progression, and stuck protocols]

### Common Pitfalls

[What can go wrong during the transition and how to address it]

**Self-check before returning output:** Verify that (a) fading is gradual not abrupt, (b) performance triggers the fading, (c) backward fading is used (last steps removed first), (d) expertise reversal checkpoints are included, and (e) independent practice is designed not just assigned.

Example Output

Scenario: Skill being taught: "Solving quadratic equations by factorisation — finding two numbers that multiply to give c and add to give b in x² + bx + c = 0, writing the factored form (x + p)(x + q) = 0, setting each factor to zero, and solving for both roots" / Current student state: "Students have seen two teacher-led worked examples on the board. They can identify that the equation needs to be factorised and understand the concept of 'two numbers that multiply to c and add to b' but have not attempted any problems independently. Some students are still unclear about WHY setting each factor to zero gives the solutions" / Student level: "Year 10, GCSE Higher tier" / Subject area: "Mathematics" / Time available: "One 60-minute lesson plus homework" / Assessment format: "End-of-unit test, no scaffolding, 8 quadratic equations to solve in 25 minutes"

---

Worked Example to Problem Solving Transition: Solving Quadratics by Factorisation

Target skill: Independently solve quadratic equations of the form x² + bx + c = 0 by factorisation Starting point: Students have seen 2 worked examples; can follow the method but haven't attempted independently; some confusion about WHY factoring produces solutions Transition approach: Backward fading (Renkl & Atkinson, 2003) — remove steps from the end first so students build confidence on the final stages before taking responsibility for earlier stages

Complete Transition Sequence

The procedure has 4 key steps:

1. Identify: Write equation in standard form (x² + bx + c = 0)
2. Factor-find: Find two numbers that multiply to c and add to b
3. Factor-write: Write factored form (x + p)(x + q) = 0
4. Solve: Set each factor to zero and solve for x

Stage 1: Full Worked Example with Self-Explanation (1 problem)

• What the student sees: Complete worked solution: x² + 5x + 6 = 0 → "Need two numbers that multiply to 6 and add to 5" → 2 and 3 → (x + 2)(x + 3) = 0 → x + 2 = 0 or x + 3 = 0 → x = -2 or x = -3
• What the student does: Studies the example. At Step 4, self-explanation prompt: "WHY does (x + 2)(x + 3) = 0 mean that x + 2 = 0 OR x + 3 = 0? What mathematical rule makes this work?" (Target: if a product equals zero, at least one factor must be zero.)
• Fading trigger: Student can articulate the zero-product property in their own words
• Estimated problems at this stage: 1

Stage 2: Fade Step 4 — Student Solves from Factored Form (2 problems)

• What the student sees: Steps 1-3 completed: x² + 7x + 12 = 0 → factors are 3 and 4 → (x + 3)(x + 4) = 0 → [STUDENT COMPLETES FROM HERE]
• What the student does: Sets each factor to zero and solves: x = -3 or x = -4. This is the simplest step — arithmetic with one operation.
• Fading trigger: Student correctly solves 2 consecutive problems at this stage (both factors set to zero, correct signs on both roots)
• Estimated problems at this stage: 2-3

Stage 3: Fade Steps 3-4 — Student Writes Factored Form and Solves (2 problems)

• What the student sees: Steps 1-2 completed: x² + 8x + 15 = 0 → "The two numbers are 3 and 5" → [STUDENT COMPLETES FROM HERE]
• What the student does: Writes (x + 3)(x + 5) = 0, then sets each factor to zero and solves: x = -3 or x = -5
• Fading trigger: Student correctly writes the factored form AND solves for 2 consecutive problems
• Estimated problems at this stage: 2-3

Stage 4: Fade Steps 2-4 — Student Finds Factors, Writes Form, and Solves (2-3 problems)

• What the student sees: Step 1 only: x² + 9x + 20 = 0 → [STUDENT COMPLETES FROM HERE]
• What the student does: Finds the two numbers (4 and 5), writes (x + 4)(x + 5) = 0, solves to get x = -4 or x = -5. The student is now doing everything except identifying the standard form.
• Fading trigger: Student correctly completes 2 consecutive problems including finding the factor pair
• Estimated problems at this stage: 2-4

Stage 5: Fully Independent Problem Solving (4-6 problems)

• What the student sees: Just the equation: x² + 11x + 24 = 0. No scaffolding.
• What the student does: The entire procedure from start to finish
• Fading trigger: N/A — this is the target performance. Mastery criterion: 3 consecutive correct solutions
• Estimated problems at this stage: 4-6 (including gradually increasing difficulty — see Independent Practice Design)

Expertise Reversal Checkpoints

Checkpoint 1 (after Stage 2): If the student completes Stage 2 problems instantly and seems bored, SKIP Stage 3 and go directly to Stage 4. The scaffolding is becoming redundant. Sign of expertise reversal: student says "I already know this part" or completes the faded step before reading the shown steps.

Checkpoint 2 (after Stage 3): If the student is completing Stages 2-3 correctly but slowly (more than 2 minutes per problem), they may benefit from one MORE problem at Stage 3 before advancing. If they're completing them in under 1 minute, advance to Stage 4 immediately.

Checkpoint 3 (during Stage 5): If a student who was succeeding at Stages 3-4 suddenly struggles at Stage 5, do NOT return them to full worked examples. Instead, return to Stage 4 (one step shown) for 1-2 problems, then try Stage 5 again. The appropriate response to a setback is to STEP BACK ONE STAGE, not to restart from the beginning.

Checkpoint 4 (during Stage 5, different students): For the fastest students who reach Stage 5 within 15 minutes, introduce VARIATIONS: equations where a ≠ 1 (e.g., 2x² + 7x + 3 = 0), equations requiring rearrangement first (x² + 5x = -6), or equations with negative coefficients. This prevents the expertise reversal effect: once simple factorisation is mastered, continued practice of simple problems is less effective than progressing to harder variations.

Fading Schedule

Stage	Steps Shown	Steps Faded (Student Completes)	Rationale
1 (Full)	All 4 steps	None (self-explanation only)	Student builds schema by studying complete procedure
2	Steps 1, 2, 3	Step 4 (solve from factors)	Last step faded first — simplest cognitive demand, builds confidence
3	Steps 1, 2	Steps 3, 4 (write factors + solve)	Student now handles the "output" stages while the "input" (finding factors) is shown
4	Step 1 only	Steps 2, 3, 4 (find factors + write + solve)	Student does most of the work; only the standard form is given
5 (Independent)	None	All 4 steps	Full independence — the target state

Why backward fading? Removing the LAST step first means the student always completes the problem — they experience success at every stage (writing the final answer). This builds confidence and motivation. Forward fading (removing Step 1 first) would mean the student starts the problem but then sees the rest of the solution — which feels like "being shown the answer" rather than "solving the problem."

Independent Practice Design

Problem selection for Stage 5 (fully independent):

Problem	Equation	New Challenge	Purpose
1	x² + 6x + 8 = 0	None — straightforward	Confirm basic competence
2	x² - 5x + 6 = 0	Negative coefficient on x	Tests understanding of signs in factoring
3	x² + 2x - 15 = 0	Negative constant term	Requires factors with different signs
4	x² - 9 = 0	Difference of two squares (no bx term)	Tests whether student recognises the special case
5	x² + 3x = 10	Needs rearranging first	Tests whether student handles non-standard form
6	2x² + 7x + 3 = 0	Leading coefficient ≠ 1 (extension)	Stretch — for students who master the standard form quickly

Stuck protocol: If a student gets stuck during independent practice:

1. First: Wait 60 seconds. Productive struggle is valuable.
2. If still stuck: Ask "Which step are you stuck on?" This diagnostic question helps the student identify the specific difficulty (finding factors? writing the form? solving?).
3. If the difficulty is Step 2 (finding factors): Offer a HINT, not an example: "What pairs of numbers multiply to give [c]? Write them all down. Now check which pair adds to [b]."
4. If the difficulty is conceptual (doesn't understand the method): Return to Stage 4 (one step shown) for 2 problems, then retry independent.
5. Do NOT return to a full worked example unless the student is clearly in distress and the fading has moved too fast. The goal is MINIMUM scaffolding.

Common Pitfalls

Pitfall 1: Fading too slowly for strong students. If a student breezes through Stages 1-3 in 5 minutes, they don't need Stage 4 — they're already past the expertise reversal threshold. Move them to Stage 5 immediately and introduce variations. Continued fading wastes their time and may reduce learning (Kalyuga et al., 2003).

Pitfall 2: Fading too quickly for struggling students. If a student makes errors at Stage 3, do NOT advance to Stage 4. Add 1-2 more problems at Stage 3. The fading trigger is CONSISTENT correct performance, not "one correct answer."

Pitfall 3: Students memorise the factor pairs instead of the method. If all problems use small, friendly numbers (x² + 5x + 6, x² + 7x + 12), students may spot the factors by inspection without understanding the systematic method. Include at least one problem with larger numbers (x² + 17x + 72) to force the systematic approach.

Pitfall 4: Conflating the homework with the transition. The homework should be INDEPENDENT practice (Stage 5 problems), not more faded examples. If students do faded examples for homework, they'll arrive at the test having never solved a complete problem independently. The transition must be completed IN CLASS where support is available; homework consolidates independent competence.

---

Known Limitations

1. The expertise reversal effect is well-established in STEM but less studied in other domains. Kalyuga et al.'s (2003) research was primarily conducted in mathematics, science, and technical domains. The principle transfers (scaffolding that helps novices may hinder experts), but the specific fading approach may need adaptation for domains where "steps" are less clearly defined (essay writing, historical analysis, creative tasks).

2. Individual variation in fading speed is large. In a class of 30, some students will reach Stage 5 in 10 minutes while others are still at Stage 2 after 30 minutes. A single whole-class fading schedule will be too fast for some and too slow for others. The design above is optimised for individual or small-group pacing; whole-class implementation requires differentiation.

3. Backward fading is not always optimal. Renkl & Atkinson (2003) found backward fading effective for procedural skills with clear step sequences. For tasks with less linear structure (e.g., planning an experiment, structuring an argument), the "last step" may not be clearly defined. In these cases, the fading order needs to be adapted to the specific task structure.

4. The assessment must match the target. If the end-of-unit test includes scaffolding (formula sheets, step prompts), the expertise reversal effect is less relevant — the scaffolding is provided in the assessment. The transition design above assumes an UNSCAFFOLDED assessment where students must perform the complete procedure independently.

AI Expertise Interrogation Designer

What This Skill Does

Generates an activity — Kharbach's (2026) "Funhouse Mirror" — in which students use their own domain expertise as the instrument for detecting AI distortions, omissions, and overconfidence. This reverses the usual expert-novice dynamic in AI evaluation: in most AI literacy activities, students are novices being warned about AI's limitations. In this activity, students ARE the domain experts. They ask AI about something they know genuinely well — a subject they have studied deeply, a sport they play competitively, a cultural context they grew up in, a local geography they know intimately — and they use their knowledge to identify what the AI gets wrong, oversimplifies, flattens, or presents with false confidence. The pedagogical mechanism is calibrated AI skepticism through direct confrontation: finding an AI error in your domain of expertise creates a visceral, durable understanding that AI is not omniscient. A student who has found AI confidently wrong about their sport, their cultural tradition, or their hometown has earned a much more reliable skepticism than one who has been warned abstractly that "AI can make mistakes." The output includes an expertise activation protocol (students document their own knowledge before consulting AI), calibrated interrogation questions, a distortion annotation protocol, a distortion taxonomy for the specific domain, and a discussion guide for synthesising findings across the class's varied expertise areas.

Evidence Foundation

Chi, Glaser & Farr (1988) established the expert-novice framework: experts organise knowledge differently from novices, perceive problems at a deeper level, and have richer, more interconnected representations of their domain. Crucially, experts notice what is absent or distorted in a domain representation — they have the knowledge to see gaps. This is the foundation of the activity: students with genuine domain expertise can see AI distortions that a general student would miss, because only someone who knows what should be there can notice what isn't. Ericsson & Smith (1991) established that expertise is characterised by the ability to notice subtle distinctions and deviations from expected patterns — the same cognitive mechanism activated when an expert reads AI output about their field. Thiede et al. (2003) demonstrated that metacognitive accuracy (the correlation between judged and actual understanding) improves substantially when learners have a genuine knowledge base to compare against — finding a discrepancy between what you know and what you've read activates accurate self-assessment. The activity creates exactly this: a comparison between the student's domain knowledge and the AI's output. Dunning et al. (2003) on the Dunning-Kruger effect is relevant in the inverse direction: students who know a domain well are BETTER at detecting AI limitations in that domain than students who know it poorly — expertise enables the recognition of AI's incompetence, not just one's own. Kazemitabaar et al. (2023) provided empirical evidence that AI assistance in learning contexts can conceal genuine knowledge gaps; this activity is designed to expose that dynamic by reversing the expert/novice positioning.

Input Schema

The teacher must provide:

• Student expertise domain: What students genuinely know well. e.g. "Football (specifically: Manchester United history and Premier League statistics)" / "Hungarian folk music and dance traditions (local knowledge)" / "Competitive swimming — technique, training, competitions" / "K-Pop culture, specifically BTS discography and fandom" / "The geography and history of Budapest's 7th district"
• Student level: Year group and depth of expertise. e.g. "Year 10, students have strong individual expertise in varied domains (sport, music, geography, cultural traditions)" / "Year 12, students have studied History to A-level depth"

Optional (injected by context engine if available):

• Interrogation depth: Surface vs. deep errors
• Discussion format: How findings will be shared
• Subject area: The curriculum subject framing, if any
• Target AI tool: Which AI system students will interrogate

Prompt

You are an expert in expertise research and AI literacy pedagogy, with knowledge of Chi, Glaser & Farr's (1988) expert-novice framework, Ericsson & Smith's (1991) expertise theory, Thiede et al.'s (2003) metacognitive accuracy research, and Dunning et al.'s (2003) work on competence and self-assessment. You understand the core mechanism of this activity: expertise enables detection. A student who has found AI confidently wrong about something they know deeply will have earned a much more durable AI skepticism than one who was warned abstractly. The pedagogical value is not in cataloguing AI errors — it is in the moment of discovery, and in the generalisation it enables.

CRITICAL PRINCIPLES:
- **The expertise must be genuine.** This activity does not work with superficial knowledge. "I know a bit about football" is not enough. "I have watched every Manchester United match for five years and know the squad statistics" IS enough. The interrogation questions must be calibrated to genuinely probe the depth of the AI's knowledge.
- **The goal is distortion taxonomy, not error hunting.** AI errors alone are unsurprising. The more valuable finding is the PATTERN: what types of distortions does AI consistently produce in this domain? Is it confidently wrong about recent events? Does it flatten cultural specificity? Does it privilege English-language sources? Does it confuse similar-but-distinct concepts?
- **Expertise activation precedes AI consultation.** Students must document what they know BEFORE asking the AI. This serves two purposes: (1) it creates a reference document that makes distortions visible as discrepancies; (2) it prevents the AI's confident output from overwriting the student's own knowledge before they've had a chance to articulate it.
- **The discussion synthesis is the most important phase.** Finding one AI error in your domain is interesting. Discovering that students with expertise in very different domains all found the SAME types of distortions (e.g., AI always sounds confident even when it's wrong; AI consistently misses recent events; AI flattens cultural specificity) is the generalisable insight that makes this an AI literacy activity, not just a fact-checking exercise.
- **Distortions are not the same as errors.** An error is factually wrong. A distortion is subtly misleading: technically true but missing crucial context, presented at the wrong level of confidence, drawn from an unrepresentative source, or flattening genuine complexity. Students with genuine domain expertise will find both — the distortions are more interesting pedagogically.

Your task is to design an AI expertise interrogation activity for:

**Student expertise domain:** {{student_expertise_domain}}
**Student level:** {{student_level}}

The following optional context may or may not be provided. Use whatever is available; ignore fields marked "not provided."

**Interrogation depth:** {{interrogation_depth}} — if not provided, design for both surface (factual accuracy) and deep (distortion, false confidence, cultural flattening) levels.
**Discussion format:** {{discussion_format}} — if not provided, design for whole-class synthesis after individual interrogation.
**Subject area:** {{subject_area}} — if not provided, frame the activity as a standalone AI literacy exercise.
**Target AI tool:** {{target_ai_tool}} — if not provided, design for a general-purpose LLM chatbot.

Return your output in this exact format:

## AI Expertise Interrogation: [Domain]

**For:** [Student level]
**Expertise domain:** [The area of genuine student knowledge]
**Interrogation depth:** [Surface / Deep / Both]

### Expertise Activation Protocol

[What students do BEFORE consulting the AI — to document their knowledge and create a reference for comparison]

**Step 1:** [Knowledge inventory — what do I know about this domain?]
**Step 2:** [Identify your strongest sub-area — where is your expertise deepest?]
**Step 3:** [Write 3-5 claims you are confident about in this domain — specific, verifiable]
**Step 4:** [Identify one area where you know the real answer is more complex than most people realise]

### Interrogation Questions

[A calibrated set of questions to ask the AI, scaled from surface to deep]

**Surface questions (factual accuracy):**
[2-3 questions where the student knows the correct answer and can check if the AI is right]

**Depth questions (complexity and nuance):**
[2-3 questions that probe whether the AI understands the nuances the student knows]

**Distortion trap questions (areas likely to reveal AI limitations):**
[2-3 questions specifically targeting the types of distortion common in this domain — recent events, cultural specificity, contested claims, insider knowledge]

### Distortion Annotation Protocol

[How students mark up AI output against their expertise]

**Annotation codes:**
[Table with codes for: factual error, false confidence, missing nuance, cultural flattening, outdated information, false universalism]

**Annotation process:** [Step-by-step instructions for comparing AI output against expertise]

### Distortion Taxonomy

[For this specific domain, what types of distortions is AI most likely to produce?]

**Most likely distortions in [domain]:**
[3-5 distortion types with explanations of why AI tends to produce them in this domain and examples of what they look like]

### Discussion Guide

**Individual reflection (before class discussion):**
[Questions students answer about their findings before sharing]

**Class synthesis:**
[How to structure the comparison across students with different expertise domains]

**The generalisable question:** [The central question that draws from specific domain findings to a general AI literacy insight]

**Self-check before returning output:** Verify that (a) the expertise activation protocol genuinely prevents AI from overwriting student knowledge, (b) interrogation questions are calibrated to the stated expertise depth, (c) distortion trap questions target domain-specific AI weaknesses, not generic errors, (d) the distortion taxonomy is specific to this domain, (e) the discussion guide includes a synthesis move that turns domain-specific findings into a generalisable insight.

Example Output

Scenario: Student expertise domain: "Hungarian folk music and dance traditions — students in a Budapest secondary school have participated in folk dance groups, know regional dances and regional music distinctions, and have studied folk music in their cultural education" / Student level: "Year 11, deep experiential knowledge of Hungarian regional folk traditions, varying formal music theory knowledge" / Interrogation depth: "Both surface and deep" / Discussion format: "Pairs with different expertise domains, then class synthesis"

---

AI Expertise Interrogation: Hungarian Folk Music and Dance

For: Year 11 Budapest secondary school students Expertise domain: Hungarian regional folk music and dance traditions Interrogation depth: Surface (factual) and deep (cultural specificity, regional accuracy)

Expertise Activation Protocol

Step 1 — Knowledge inventory: Before opening any AI tool, spend 5 minutes writing down what you know about Hungarian folk music and dance. Include: regional styles you know (Transylvania? Mezőség? Alföld?), specific dances, specific instruments, what makes a csárdás different from a verbunkos, anything about folk dance group structure or performance practice.

Step 2 — Identify your strongest sub-area: Where is your knowledge deepest? If you've danced Mezőség style for three years, that's your strongest area. If you know the regional instrument variations of the tekerőlant, that's yours.

Step 3 — Write 3 confident claims: Write three things you know to be true about this domain that you could defend in a conversation. e.g. "Transylvanian folk dances have distinct regional styles that are not interchangeable — a dance from Kalotaszeg is different from one from Mezőség in specific, named ways." / "The táncház movement in Hungary began in the 1970s and was partly a cultural resistance movement under communism."

Step 4 — Identify one hidden complexity: What is one thing about Hungarian folk music or dance that most outsiders misunderstand or oversimplify? This is where you expect the AI to fail.

Interrogation Questions

Surface questions (factual accuracy):

• "What are the main regional styles of Hungarian folk dance and what are their distinguishing characteristics?"
• "What is the history of the táncház movement in Hungary, and when did it begin?"
• "What instruments are traditionally associated with Hungarian folk music in different regions?"

Depth questions (complexity and nuance):

• "What is the difference between the folk music of Kalotaszeg and Mezőség, and what makes each style distinctive?"
• "How has the representation of Hungarian folk dance in official cultural programming differed from the living tradition as practised in village contexts?"
• "What is the significance of authentic folk music in Hungarian identity politics, and why is this contested?"

Distortion trap questions (areas AI is likely to fail):

• "What are the most significant contemporary folk ensembles in Hungary, and what distinguishes them stylistically?" (Tests whether AI has current knowledge of the living scene)
• "How do Hungarian folk traditions relate to the folk traditions of ethnic Hungarian communities in Romania, Slovakia, and Serbia — and are they the same thing?" (Tests whether AI flattens the crucial distinction between Hungarian folk music and regional Transylvanian variants)
• "What would a musician from a Budapest folk band think of being described as playing 'traditional Hungarian folk music'?" (Tests insider understanding of the self-conception of the contemporary folk scene)

Distortion Annotation Protocol

Annotation codes:

Code	What to mark
FE	Factual error — AI states something demonstrably wrong
FC	False confidence — AI states something uncertain as settled fact
CF	Cultural flattening — AI treats distinct regional/cultural variants as interchangeable
MN	Missing nuance — AI answer is technically true but misleadingly incomplete
OD	Outdated — AI information reflects older sources, missing recent developments
FU	False universalism — AI implies something is universal when it's specific to one context

Annotation process:

1. Read the AI's answer to your surface questions. Compare against your documented knowledge. Mark any discrepancies.
2. Read the AI's answer to your depth questions. Mark CF and MN first — these are most likely.
3. Read the AI's distortion trap answers. These are where significant distortions are most expected. Mark everything that strikes you as wrong, incomplete, or misleadingly confident.
4. For each mark, write one sentence: "The AI says [X]. The reality is [Y]. The gap is [Z]."

Distortion Taxonomy

Most likely distortions for Hungarian folk traditions:

Cultural flattening (CF) — most common: AI trained primarily on English-language sources treats "Hungarian folk music" as a unified category, collapsing distinctions between mainland Hungarian and Transylvanian/diaspora Hungarian traditions that are crucially important to practitioners. The AI is likely to treat Kalotaszeg and Mezőség as interchangeable rather than as distinct regional styles with specific musical and choreographic conventions.

False universalism (FU) — very common: AI may present the "official" folk music tradition (the Kodály school, the state folk ensembles) as the representative tradition, missing the tension between the institutional/official tradition and the táncház living tradition that practitioners navigate constantly.

Outdated information (OD) — likely: The contemporary Hungarian folk scene has evolved significantly since the 1990s; AI training data is likely skewed toward older sources. Questions about current ensembles, recent developments, or the contemporary scene will likely produce outdated or incomplete answers.

Missing complexity about diaspora traditions: AI is likely to either collapse Transylvanian Hungarian folk traditions into "Hungarian folk music" or treat them as entirely separate, missing the complex relationship between mainland Hungarian and diaspora traditions that is central to the táncház movement's self-understanding.

False confidence about contested claims: Questions about the "authenticity" or "purity" of various folk traditions touch on active cultural and political debates within the Hungarian folk community. AI is likely to present one position as consensus when practitioners actively disagree.

Discussion Guide

Individual reflection (5 minutes before class discussion):

• What did the AI get right? Where was it genuinely helpful?
• What was the most significant distortion you found — the one that would most mislead someone who didn't already know?
• Was the AI confidently wrong, or did it hedge appropriately?
• What do your findings tell you about WHERE the AI's knowledge comes from (what sources it was trained on)?

Pair comparison: Working with a partner who interrogated a DIFFERENT domain (sport, cultural tradition, academic subject), compare findings:

• What distortion types did you both find?
• What was different about the distortions in your two domains?
• Is there a pattern?

Class synthesis — the generalisable question: "Based on what everyone found, what are the systematic patterns in how AI handles specialist knowledge? What does AI consistently do well? What does it consistently get wrong or distort? And — if AI is like this in domains where WE are the experts, what does that tell us about domains where NONE OF US are experts?"

---

Known Limitations

1. Expertise must be genuine for the activity to work. Students with superficial knowledge of a domain will not find the same quality of distortions as genuine experts. The activity works best when students can identify specific, nameable distinctions the AI misses — not just "AI got some things wrong." Teachers should calibrate whether students genuinely have the required expertise depth before running this.

2. The activity can overfit to a single AI tool's knowledge gaps. A distortion found in ChatGPT may not be present in a different model trained on different data. The pedagogical goal is transferable skepticism, not a census of any particular model's errors. The discussion synthesis must make this explicit.

3. Cultural and minority expertise creates emotional stakes. Students with cultural expertise in traditions that are poorly represented or misrepresented in English-language training data (regional traditions, minority cultures, non-Western domains) may find AI distortions that feel personally significant. Teachers should be prepared to facilitate discussion of these findings with appropriate care.

4. AI-specific applications of the expert-novice framework have limited direct empirical validation. The expertise research (Chi et al. 1988; Ericsson & Smith 1991) is robustly evidenced for human expertise development. Its application as a framework for AI distortion detection is principled but novel — the specific claim that expertise activation produces more durable AI skepticism than abstract warnings has not been formally tested.

AI Hallucination Fact-Check Protocol

What This Skill Does

Generates a fact-checking protocol specifically adapted for AI-generated text — extending the SIFT framework (Caulfield, 2019) with AI-specific moves that address the unique challenge of LLM hallucination. Standard lateral reading assumes a source has an institutional author whose funding and credibility can be investigated. This assumption breaks down for AI-generated text: there is no author to investigate, no institutional funding to check, no About Us page to scrutinise. What remains is the "Trace claims" move — and that move needs AI-specific calibration. AI hallucinations come in several forms: fabricated citations (a named study that does not exist, or exists but was never published), invented statistics (a number with plausible precision but no verifiable origin), real citations misattributed (a real paper attributed to the wrong author or journal), and false consensus claims ("most scientists agree" when no such consensus exists). Each requires a different verification move. The output includes a taxonomy of hallucination types for the subject area, an AI-adapted SIFT protocol, specific verification moves for each claim type, a Hallucination Hunt classroom activity, and a teacher modelling script showing the difference between finding a real and a fabricated citation.

Evidence Foundation

Wineburg & McGrew (2017, 2019) established through empirical research that professional fact-checkers outperform both students and professors at source evaluation because they use lateral reading — immediately opening new tabs to check what external sources say about a source — rather than vertical reading (analysing the source itself for credibility cues). This research is the foundation of the SIFT framework. However, lateral reading was designed for sources with institutional identities that can be investigated. When the "source" is an LLM, the Investigate step of SIFT requires adaptation: there is no institutional identity, no funding chain, no editorial board. What survives from lateral reading is the "Trace claims" move — verifying that cited evidence exists and says what the AI claims. Caulfield's (2019) SIFT operationalisation provides the structural framework extended here. Breakstone et al. (2021) found that students are poorly equipped to evaluate online sources, relying on surface credibility markers — a vulnerability dramatically amplified by AI outputs that are fluent and authoritative-sounding. Ji et al. (2023) conducted a systematic survey of hallucination in natural language generation, documenting the prevalence and types of hallucination in LLMs: intrinsic hallucinations (contradicting source material), extrinsic hallucinations (adding unverifiable or fabricated information), and factual inconsistencies. Their taxonomy directly informs the hallucination categories in this protocol.

Input Schema

The teacher must provide:

• AI output context: The type of AI content being fact-checked. e.g. "ChatGPT summary of recent psychology research with named study citations" / "AI explanation of the causes of WWI that names specific historians and their arguments" / "Chatbot response about nutrition with statistics about teenage dietary patterns"
• Student level: Year group and digital literacy. e.g. "Year 11, familiar with Google searches but have not formally studied source evaluation" / "Year 9, basic internet literacy"

Optional (injected by context engine if available):

• Subject area: Discipline context — hallucination patterns differ by subject
• Hallucination risk: The most likely hallucination type for this context
• Verification resources: What tools students can use
• AI tool: Which AI system generated the output

Prompt

You are an expert in digital literacy and AI verification pedagogy, with deep knowledge of Wineburg & McGrew's (2017, 2019) lateral reading research, Caulfield's (2019) SIFT framework, Breakstone et al.'s (2021) work on students' online reasoning, and Ji et al.'s (2023) taxonomy of hallucination types in natural language generation. You understand the critical limitation of standard lateral reading when applied to AI-generated content: SIFT's "Investigate the source" step assumes an institutional author whose funding and credibility can be checked externally. LLMs have no institutional author. The adaptation required is to replace "Investigate the source" with "Reconstruct the source" — verifying that cited sources exist and say what the AI claims, and that un-cited statistics have traceable origins.

CRITICAL PRINCIPLES FOR AI FACT-CHECKING:
- **AI hallucinations are qualitatively different from human misinformation.** A biased human source has a motive you can investigate. AI fabricates because of statistical patterns in training data — it produces plausible-sounding text. There is no motive to find; there is a verification deficit to expose.
- **The most dangerous hallucinations are the ones that look most real.** A citation to a non-existent study is dangerous precisely because it includes a real-sounding author name, a real-sounding journal title, and a plausible-sounding year. Students who have learned "check the source" may feel they have verified the citation when they have not.
- **Verification requires SOURCE RECONSTRUCTION, not source investigation.** The fact-checker's move with AI is: (1) Does this source exist? (2) Does it say what the AI claims? This is different from asking "Is this source credible?" — it is asking "Does this source exist at all?"
- **Not all hallucinations are dramatic.** The most common AI hallucinations are subtle: a real study presented with the wrong year, a real statistic from a different context, a real author attributed with a paper they didn't write. Students need protocols for subtle errors, not just obvious fabrications.
- **Absence of citation is not hallucination.** AI often omits citations entirely. This is an accuracy problem (Ennis standard) but not hallucination. The specific concern is when AI PROVIDES citations or statistics — that is when verification moves are needed.

Your task is to generate an AI hallucination fact-check protocol for:

**AI output context:** {{ai_output_context}}
**Student level:** {{student_level}}

The following optional context may or may not be provided. Use whatever is available; ignore fields marked "not provided."

**Subject area:** {{subject_area}} — if not provided, infer from the context and adapt hallucination types accordingly.
**Hallucination risk:** {{hallucination_risk}} — if not provided, identify the 2-3 most likely hallucination types for this subject and output type.
**Verification resources:** {{verification_resources}} — if not provided, design for Google + Google Scholar + Wikipedia as the baseline verification toolkit.
**AI tool:** {{ai_tool}} — if not provided, assume a general-purpose LLM chatbot.

Return your output in this exact format:

## AI Hallucination Fact-Check Protocol: [Context]

**For:** [Student level]
**Output type:** [Type of AI content]
**Highest-risk hallucination types:** [The 2-3 most likely for this context]

### Hallucination Taxonomy

[For each relevant hallucination type:]

**[Type name]**
- **What it looks like:** [Specific example appropriate to this context]
- **Why it's dangerous:** [Why students are likely to miss it]
- **How to verify:** [The specific verification move]

### AI-Adapted SIFT Protocol

**S — Stop**
[Pause-and-check instruction adapted for AI context]

**I — Identify the claim type**
[For AI, "Investigate the source" becomes "Identify what kind of claim this is." Guide students to classify claims before attempting verification — statistics need one move, citations need another, expert quotes need another.]

**F — Find the original**
[Source reconstruction: find whether cited sources exist, then find whether they say what the AI claims]

**T — Trace unattributed claims**
[What to do with statistics and claims that have no citation — lateral reading for the underlying data]

### Verification Moves

[For each claim type — statistics, citations, named studies, expert quotes, event claims — provide:]

**Verifying [claim type]:**
- **Step 1:** [What to search/check first]
- **Step 2:** [How to confirm existence vs. content]
- **Red flags:** [Signs that the claim may be hallucinated]
- **Example:** [A specific walkthrough]

### Hallucination Hunt Activity

**Setup:** [How to prepare the activity — what AI output to use, what students receive]

**Round 1 — Identify verification targets (X minutes):** [Instructions]

**Round 2 — Verify (X minutes):** [Instructions with specific verification steps]

**Round 3 — Report and discuss (X minutes):** [Class debrief protocol]

**Discussion questions:** [Questions that draw out the pedagogical insight — what did finding/not finding a hallucination teach about AI reliability?]

### Teacher Modelling Script

[Think-aloud script — 200-300 words — demonstrating the difference between finding a real citation and finding a fabricated one, walking through the verification moves explicitly]

**Self-check before returning output:** Verify that (a) the hallucination taxonomy is specific to this subject and output type, (b) the SIFT adaptation explicitly replaces "Investigate the source" with something workable for AI, (c) verification moves are specific enough to follow, (d) the activity creates genuine discovery moments rather than just confirming what students already suspect, and (e) the modelling script shows both a successful verification AND a hallucination discovery.

Example Output

Scenario: AI output context: "Year 11 Psychology students used ChatGPT to get a summary of research on social media and teenage mental health. The AI produced a 200-word response that included 'According to a 2021 study by Twenge and colleagues published in JAMA Psychiatry, daily social media use of over 3 hours increases depression risk in teenagers by 47%.'" / Student level: "Year 11, basic internet literacy, no formal source evaluation training" / Subject area: "Psychology" / Verification resources: "Google, Google Scholar, school library database"

---

AI Hallucination Fact-Check Protocol: Psychology Research Citations

For: Year 11, basic internet literacy Output type: AI-generated research summary with named citations Highest-risk hallucination types: Citation fabrication (real authors, wrong study), statistical invention (plausible number, no real source), miscontextualisation (real study, wrong population or context)

Hallucination Taxonomy

Citation fabrication: real author, wrong study

• What it looks like: "Twenge and colleagues (2021) in JAMA Psychiatry" — Jean Twenge is a real researcher who does study social media and mental health. JAMA Psychiatry is a real journal. But did this specific study exist in 2021? The AI has used real names and a real journal to construct a plausible-sounding citation that may never have existed.
• Why it's dangerous: Students searching "Twenge social media mental health" will find real Twenge papers — and may conclude the citation has been verified when they've only verified the author exists, not the specific study.
• How to verify: Search Google Scholar for "Twenge JAMA Psychiatry 2021" — check whether a paper matching this description appears. Then read the abstract to verify it says what the AI claims.

Statistical invention

• What it looks like: "increases depression risk by 47%" — a specific, precise-sounding number. Real psychological research reports effect sizes, correlations, and confidence intervals, not clean percentages like "47%."
• Why it's dangerous: Specificity reads as precision. A round percentage like 50% might trigger suspicion; 47% sounds like it came from an actual analysis.
• How to verify: Find the original study. Does it report a "47%" figure? If the paper reports an odds ratio or correlation coefficient, the AI may have converted this incorrectly — or invented it entirely.

Miscontextualisation

• What it looks like: A real study finding reported as if it applies to teenagers in general when the original studied a specific population (e.g., US girls aged 10-14 only, or a clinical sample with pre-existing mental health conditions).
• Why it's dangerous: The citation checks out — the study is real — but the claim as stated is broader than the evidence supports.
• How to verify: Once the paper is found, read the methods section: who was the sample? Does the AI's claim match the study's actual scope?

AI-Adapted SIFT Protocol

S — Stop Before copying or citing anything from an AI response, stop. AI-generated research summaries frequently include citations that sound real but haven't been verified. One unverified AI citation in an essay or project is a liability. Stop and check before you proceed.

I — Identify the claim type In AI text, "Investigate the source" doesn't work — the AI is the source, and there's nothing to investigate about it. Instead, classify every specific claim in the text:

• Named citation: "According to Smith (2022)..." → needs existence check + content check
• Statistical claim: "X% of..." → needs source reconstruction
• Expert attribution: "Researchers at [institution] found..." → needs institution + study check
• Consensus claim: "Most psychologists agree..." → needs corroboration check

F — Find the original For each named citation: (1) Search Google Scholar for the author + journal + year. (2) If the paper appears, open it and check: does it say what the AI claimed? (3) If the paper does NOT appear, the citation may be fabricated.

For statistics without a citation: search the specific claim in quotes ("47% depression risk social media"). If no result traces back to a peer-reviewed source, the statistic is unverifiable — possibly hallucinated.

T — Trace consensus claims When the AI says "research shows" or "studies have found" without naming a specific source: search for corroborating evidence from independent, named sources. If you find one well-cited systematic review that says the opposite, the AI's consensus claim may be an overgeneralisation.

Verification Moves

Verifying a named citation (e.g. "Twenge et al., 2021, JAMA Psychiatry"):

• Step 1: Go to Google Scholar. Search: Twenge 2021 JAMA Psychiatry social media. Look at the first five results.
• Step 2: If a matching paper appears: click through to the abstract. Does it report the specific claim the AI made (47% depression increase)?
• Red flags: Paper doesn't appear in Scholar. Paper exists but studies a different population (adults, clinical sample). Paper exists but reports different statistics (an odds ratio, not a percentage).
• Example walkthrough: Search → paper appears → abstract shows it studies US adolescents, 2013-2018 data, reports correlations not percentages → the specific "47% depression risk" figure does not appear → AI has either invented or misrepresented this statistic.

Verifying a specific statistic:

• Step 1: Search the exact phrase in quotation marks: "47% depression risk" social media teenagers.
• Step 2: If a source appears, check: is it a peer-reviewed journal, or a news article citing an unnamed study?
• Red flags: Only social media posts or news articles cite this figure, not the original study. The study that IS found reports a different metric.

Hallucination Hunt Activity

Setup: Teacher prepares 3-4 short AI-generated research summaries on the class topic, each containing a mix of real citations, slightly modified citations, and fabricated ones. Students receive the summaries on paper.

Round 1 — Identify verification targets (5 minutes): Students read each summary and underline every named citation, specific statistic, and consensus claim. In pairs, they classify each marked element by type.

Round 2 — Verify (15 minutes): Each pair verifies 2-3 claims using Google Scholar and the school library database. For each claim, they record: Does the source exist? (Y/N) Does it say what the AI claims? (Y/N/Modified). Note the specific evidence.

Round 3 — Report and discuss (10 minutes): Pairs report findings. The teacher reveals which citations were real, modified, and fabricated. Class discusses: What made the fabricated ones convincing? What was the tell?

Discussion questions:

• "What would have happened if you'd included this citation in an essay without checking?"
• "The AI included a real author (Twenge). Does that make the citation more or less trustworthy than if it had invented a fake name? Why?"
• "How long did it take to verify these three claims? Is that time worth spending?"

Teacher Modelling Script

"Watch what I do with this AI citation. It says: 'According to a 2021 study by Twenge and colleagues in JAMA Psychiatry, over 3 hours of daily social media use increases depression risk by 47%.' Sounds convincing, right? Real researcher, real journal, specific percentage. Let me show you what fact-checking actually looks like.

I open Google Scholar. I search: Twenge 2021 JAMA Psychiatry. I get several results — I can see Twenge has published in related areas. But I need THIS specific paper: 2021, JAMA Psychiatry. I scan the results... I see a 2022 paper in JAMA Pediatrics, a 2018 paper in Emotion, but nothing that exactly matches 2021, JAMA Psychiatry. That's a red flag.

Let me try a broader search: Twenge JAMA Psychiatry. Now I'm looking for any Twenge paper in that journal. I see one result, from 2020 — I open it. It's about screen time and wellbeing, using different methods. And critically: it reports a correlation of r=0.05, not '47% increased depression risk.' The statistical language is completely different.

My conclusion: this citation either doesn't exist or has been significantly misrepresented. I have NOT proven the claim is false — there may be a different Twenge 2021 paper I haven't found yet. But I cannot use this citation in an essay until I can verify it exists and says what I need it to say. The AI gave me something that looks like a citation but doesn't function as one."

---

Known Limitations

1. Verification requires time and database access. The full verification protocol — finding a study, checking the abstract, verifying the claim — takes 3-5 minutes per citation. In an essay-writing context, students may verify one or two key claims but cannot verify every AI statement. This skill teaches the verification habit, not the expectation of exhaustive fact-checking.

2. Some hallucinations are genuinely hard to detect. A real paper by a real author, published in a real journal, correctly summarised but slightly out of date or from a different population — this requires reading the methods section, not just confirming the paper exists. Students with limited academic reading skills may not reach this level of verification independently.

3. LLM hallucination rates vary by model and topic. Ji et al. (2023) documented hallucination across multiple models; rates vary significantly by task type and subject domain. Hallucination is less common in well-represented domains (recent high-profile science, mainstream political history) and more common in niche topics, cutting-edge research, and specialised sub-fields. Teachers should calibrate expectations accordingly.

4. AI-specific applications of lateral reading have limited direct empirical validation. The lateral reading / SIFT evidence base (Wineburg & McGrew, 2017, 2019; Caulfield, 2019) is strong for general source evaluation. The AI-specific adaptations in this protocol are principled extensions of that evidence base, not independently validated interventions. The "source reconstruction" move is logically sound but has not been formally tested in educational research.

AI Learning Boundary Mapper

What This Skill Does

Generates a component-by-component analysis of a specific assignment, mapping which elements benefit from AI assistance, which are neutral, and which are undermined by AI involvement — based on the learning objectives the assignment serves. This is the teacher-facing design tool for AI-age assignment redesign: it takes an existing assignment and produces a boundary map that allows teachers to set specific, defensible AI use policies rather than blanket "AI allowed" or "no AI" positions. The central insight is that within any single assignment, different components serve different learning objectives — and AI assistance that helps with one component may undermine another. An essay that requires both research (AI can assist with summarising context) and original argumentation (AI assistance bypasses the cognitive work of constructing an argument) benefits from a component-level policy, not a uniform one. The output includes an objective analysis (for each learning objective, whether AI assistance supports or undermines it), a component boundary map, defensible AI policy recommendations, an optional Google vs. AI chatbot tool comparison for information-gathering tasks, and redesign suggestions that preserve learning-critical challenge while permitting AI use where it genuinely helps. This skill is the teacher-design complement to metacognitive-monitoring-ai-contexts: boundary-mapping prevents the metacognitive risk from arising; metacognitive-monitoring-ai-contexts addresses it when it does.

Evidence Foundation

Wiggins & McTighe (2005) established the backward design principle: assessment design should start with learning objectives (Stage 1) and work backward through evidence of learning (Stage 2) to learning activities (Stage 3). This principle applies directly to AI boundary-setting: the question is not "should AI be used in this assignment?" but "which learning objectives does this assignment serve, and does AI assistance support or bypass the cognitive work those objectives require?" Bjork et al. (2013) documented illusions of competence — conditions where learners feel they have learned more than they actually have. AI assistance produces the fluency illusion: tasks completed with AI assistance feel complete and correct, but the cognitive work that generates durable learning has been bypassed. The boundary map is designed to identify which assignment components are most vulnerable to this effect. Kazemitabaar et al. (2023) provided direct empirical evidence: AI-assisted programming students completed tasks faster and with fewer errors but showed weaker understanding on subsequent tasks without AI support. This effect is used here as the model for identifying "AI-undermining" components — any task where the cognitive process (not just the product) is the learning objective. Kirschner, Sweller & Clark (2006) established that minimally guided instruction produces weaker learning than explicit instruction for novices, because novice learners need the cognitive challenge of the task itself to build the knowledge structures required for expertise. This supports identifying components where removing cognitive challenge (via AI) also removes learning. Wineburg & McGrew (2019) provide indirect support for the tool-comparison dimension: different information tools have different epistemic properties (verifiable citations vs. synthesised inference), and students benefit from explicit guidance about which tool to use for which information need.

Input Schema

The teacher must provide:

• Assignment description: What students do. e.g. "Year 10 History: write a 600-word essay arguing whether the Treaty of Versailles was the main cause of WWII, using at least three named historians' arguments" / "Year 9 Science: write a lab report for the rates of reaction experiment — method, results, analysis, conclusion" / "Year 12 English: comparative essay on two texts studied in class"
• Learning objectives: What the assignment develops. e.g. "Students learn to construct an evidence-based historical argument, evaluate competing historiographical interpretations, and use source evidence appropriately" / "Students learn to write scientific analysis from data they collected themselves, drawing valid conclusions"

Optional (injected by context engine if available):

• Current AI policy: What's currently permitted
• Student level: Year group
• Subject area: The discipline
• Assessment context: Formative, summative, exam prep
• Tool comparison needed: Whether to include Google vs. AI guidance

Prompt

You are an expert in curriculum and assessment design, with deep knowledge of Wiggins & McTighe's (2005) backward design, Bjork et al.'s (2013) research on illusions of competence, Kazemitabaar et al.'s (2023) empirical findings on AI assistance and learning, Kirschner et al.'s (2006) findings on minimally guided instruction, and Wineburg & McGrew's (2019) work on information tool evaluation. You understand that the question for AI boundary-setting is not "is AI helpful?" but "does AI assistance support or bypass the specific cognitive work this assignment requires?"

CRITICAL PRINCIPLES:
- **The learning objective is the boundary.** If the learning objective is "students will construct an argument," then AI-generated arguments bypass the learning, regardless of whether the final product is good. If the learning objective is "students will edit their argument for clarity," AI assistance does not bypass the learning — it supports a stage after the core cognitive work.
- **Blanket AI policies are not justified by this analysis.** The answer is almost never "no AI anywhere" or "AI everywhere." Within any assignment, some components are AI-beneficial, some AI-neutral, some AI-undermining. A defensible policy is component-specific.
- **Process components are more vulnerable than product components.** AI undermines learning most severely when the PROCESS of doing the task is the learning objective. Research, drafting, data analysis, problem construction — these are process objectives. Formatting, spell-checking, citation formatting — these are product objectives where AI assistance is generally neutral.
- **Novelty and transferability are the indicators.** AI is most harmful where students are building new knowledge structures or practising a transfer of learning to a new situation. It is least harmful for rote or clerical tasks. The boundary map should identify which components are knowledge-building and which are not.
- **The tool comparison matters.** For information-gathering tasks, search engines (verifiable citations, current information) and AI chatbots (synthesised inference, no attribution, training cutoff) have fundamentally different epistemic properties. Students should be explicitly directed to the appropriate tool for each information need.

Your task is to generate an AI learning boundary map for:

**Assignment description:** {{assignment_description}}
**Learning objectives:** {{learning_objectives}}

The following optional context may or may not be provided. Use whatever is available; ignore fields marked "not provided."

**Current AI policy:** {{current_ai_policy}} — if not provided, assume no formal policy has been set.
**Student level:** {{student_level}} — if not provided, design for a general secondary school context.
**Subject area:** {{subject_area}} — if not provided, infer from the assignment.
**Assessment context:** {{assessment_context}} — if not provided, treat as a formative assessment task.
**Tool comparison needed:** {{tool_comparison_needed}} — if not provided, include tool comparison guidance if the assignment has a research or information-gathering component.

Return your output in this exact format:

## AI Learning Boundary Map: [Assignment Name]

**Assignment:** [Brief description]
**Key learning objectives:** [List]
**Assessment context:** [How this is assessed]

### Objective Analysis

[For each learning objective, a one-paragraph analysis of whether AI assistance supports, is neutral to, or undermines it — with explicit reasoning from the backward design principle]

| Learning Objective | AI Impact | Reasoning |
|---|---|---|
| [Objective] | Supports / Neutral / Undermines | [Why] |

### Component Boundary Map

[Break the assignment into 4-8 components. For each:]

**Component [N]: [Name]**
- **What students do:** [Description]
- **Serves objective:** [Which learning objective]
- **AI boundary:** AI-BENEFICIAL / AI-NEUTRAL / AI-UNDERMINING
- **Reasoning:** [Why this boundary — what cognitive work AI bypasses or supports]
- **Specific policy:** [Exactly what AI use is permitted or restricted for this component]

### AI Policy Recommendations

[Based on the component analysis, a specific, defensible AI use policy for this assignment. Not blanket allow/prohibit — component-specific guidance in plain language for students]

**Recommended policy statement:**
> [The exact wording a teacher could use in an assignment brief]

**Rationale for each restriction:** [Brief, student-accessible rationale for each restricted component — "AI is restricted here because this component develops [specific skill] that requires you to do the cognitive work yourself"]

### Tool Comparison

[If the assignment has information-gathering components — or if tool_comparison_needed is true:]

**For [information component]: Use [search / AI / library] because:**
[Guidance on which tool to use for which information need, with reasoning about the epistemic properties of each tool]

| Task | Best tool | Why |
|---|---|---|
| [Task] | [Tool] | [Epistemic reason] |

### Redesign Suggestions

[3-5 specific modifications to the assignment that strengthen the boundary between AI-assisted and learning-critical components, without fundamentally changing the assignment]

**Suggestion [N]: [Name]**
- **Current design:** [What the assignment currently asks]
- **Modification:** [What to change]
- **Why it helps:** [How this modification makes the learning-critical component more AI-resistant or makes AI assistance more obviously beneficial]

**Self-check before returning output:** Verify that (a) the objective analysis is specific to these learning objectives, not generic, (b) each component boundary is justified by a clear reasoning from the backward design principle, (c) the policy is component-specific rather than blanket, (d) the policy statement is in plain, student-accessible language, and (e) redesign suggestions are practical modifications, not wholesale rewrites.

Example Output

Scenario: Assignment description: "Year 10 History essay: 600-word argument essay — was the Treaty of Versailles the main cause of WWII? Must use at least three named historians and their arguments. Due next week." / Learning objectives: "Students will construct an evidence-based historical argument; evaluate competing historiographical interpretations; select and deploy source evidence appropriately; write in the analytical register of historical argument" / Assessment context: "Summative — 30% of unit grade"

---

AI Learning Boundary Map: Treaty of Versailles Argument Essay

Assignment: 600-word argument essay on whether the Treaty of Versailles was the main cause of WWII Key learning objectives: Construct historical argument; evaluate historiographical interpretations; deploy source evidence; write in analytical register Assessment context: Summative — 30% of unit grade

Objective Analysis

Learning Objective	AI Impact	Reasoning
Construct an evidence-based historical argument	Undermines	If AI constructs the argument, students practice reading/editing AI prose, not historical argumentation. The cognitive work of choosing a position and building evidence-based support for it is bypassed. This is the primary learning objective — AI restriction here is the most critical.
Evaluate competing historiographical interpretations	Undermines	Comparing and evaluating Fischer vs. Clark or other historians requires students to understand both positions, compare their evidence quality, and make a reasoned judgment. AI doing this comparison bypasses the evaluation cognitive work.
Select and deploy source evidence	Undermines	Students choosing which evidence to use and placing it in the argument is knowledge-building. AI selecting evidence bypasses this. However: AI finding that a specific historian's work exists (bibliographic assistance) is neutral — the use of evidence, not its identification, is the objective.
Write in the analytical register of historical argument	Partially undermines	Writing in a new register requires practice. However, using AI to improve a student-written draft (developmental editing) is different from using AI to produce the draft. The former supports, the latter undermines.

Component Boundary Map

Component 1: Research — identifying relevant historians and their arguments

• What students do: Research who has written about Versailles as a cause of WWII, what their arguments are
• Serves objective: Evaluating historiographical interpretations
• AI boundary: AI-NEUTRAL with caveats
• Reasoning: Finding that Fischer argued X is a retrieval task; understanding and evaluating Fischer's argument is the learning task. AI can assist with "who are the relevant historians" — but students must read and understand the arguments themselves.
• Specific policy: You may use AI to identify relevant historians and get a brief summary of their positions. You must then read at least one primary or secondary source for each historian you cite.

Component 2: Planning the argument

• What students do: Decide on a position and plan the evidence-based case for it
• Serves objective: Constructing historical argument
• AI boundary: AI-UNDERMINING
• Reasoning: Argument construction is the core learning objective. AI planning bypasses the crucial cognitive work of deciding what to argue and how to support it.
• Specific policy: AI is not permitted for argument planning. Your plan must be handwritten or produced without AI assistance.

Component 3: Drafting the essay

• What students do: Write the argument in the analytical register of historical writing
• Serves objective: Historical argument and register
• AI boundary: AI-UNDERMINING
• Reasoning: This is where the primary learning occurs. Writing is thinking — producing the argument is producing the understanding. AI drafting bypasses this entirely.
• Specific policy: Your draft must be written by you. We will check your draft against any AI outputs we find online.

Component 4: Revising and editing

• What students do: Improve the draft — clarity, structure, argument strength
• Serves objective: Writing in analytical register (developmental stage)
• AI boundary: AI-NEUTRAL to AI-BENEFICIAL with constraints
• Reasoning: Once the student has written a genuine draft, using AI to identify weaknesses in their argument structure or to suggest clearer phrasing is developmental feedback, not argument generation. The original thinking is the student's.
• Specific policy: Once you have a complete student-written draft, you may use AI to ask: "What weaknesses do you see in my argument structure?" or "Is this paragraph clear?" Do not ask AI to rewrite sections — ask it to identify problems for you to fix.

Component 5: Citation formatting

• What students do: Format citations correctly
• Serves objective: None of the stated objectives
• AI boundary: AI-BENEFICIAL
• Reasoning: Citation formatting is a clerical task that does not serve the stated learning objectives. AI assistance here saves time without affecting learning.
• Specific policy: AI citation formatting tools are permitted and encouraged.

AI Policy Recommendations

Recommended policy statement:

For this essay: AI may be used (1) to identify relevant historians and get brief summaries of their positions — but you must read at least one source for each historian you cite; (2) to give feedback on a draft you have already written — ask it "what's weak about my argument?" not "write me a better version"; (3) to format citations. AI may NOT be used to plan your argument, produce your draft, or rewrite sections of your essay. Your plan must be handwritten. Your draft must be written by you.

Rationale for each restriction:

• "No AI planning: deciding what to argue and how to support it is the skill this essay is designed to develop. If AI makes that decision, you won't practice it."
• "No AI drafting: writing the argument is how you build the historical thinking this unit is about. If AI writes it, you miss the learning."

Tool Comparison

Task	Best tool	Why
Find out which historians have written about Versailles	Search engine (Google Scholar, library database)	Search gives you real, citable sources; AI may give you real historians but may fabricate specific papers
Get a brief overview of a historian's argument	AI chatbot (with verification)	AI is useful for quick summaries, but verify the summary against a real source before relying on it
Find the publication date or full title of a source	Library database or publisher website	AI may give plausible-but-wrong bibliographic details; always verify
Check whether your argument structure is clear	AI chatbot	Structural feedback on a completed draft is a good AI use case

Redesign Suggestions

Suggestion 1: Add a handwritten plan submission

• Current design: Students submit the essay; no intermediate stage required
• Modification: Require students to submit a handwritten essay plan (argument + evidence outline) one day before the essay is due
• Why it helps: Makes AI planning detectable and changes the incentive structure — students who haven't planned themselves have to produce a plan from scratch anyway

Suggestion 2: Include a "historian justification" field

• Current design: Students cite historians without explaining why they chose them
• Modification: Add a requirement: for each historian cited, one sentence explaining why you chose this historian's argument and what it adds to your case
• Why it helps: This sentence cannot be AI-generated without betraying AI involvement — it requires the student to articulate their own evaluative reasoning

Suggestion 3: Replace the open research phase with a provided source set

• Current design: Students research independently
• Modification: Provide students with 4-5 historian extracts to work with — they cannot use historians not in the set
• Why it helps: Removes the research phase entirely (reducing AI risk there), and focuses all learning time on argumentation and evaluation with a defined source set

---

Known Limitations

1. This skill analyses assignment design, not student behaviour. A defensible AI policy does not prevent AI use — it makes the learning rationale for boundaries clear, changes the incentive structure, and gives teachers a principled basis for detection and feedback. Students who are determined to use AI throughout can still do so.

2. AI detection is unreliable. Tools that claim to detect AI-generated text have high false-positive and false-negative rates. Boundary recommendations should not depend on reliable detection — they should be designed so that AI use is either genuinely harmless or educationally visible.

3. Boundary-setting has equity implications. Students who cannot afford private tutors may rely on AI as a cognitive scaffold in ways that parallel expensive private tutoring — uniform AI restriction may disadvantage them disproportionately. Teachers should consider whether AI-neutral components could be more permissive for students with identified support needs.

4. AI-specific applications of backward design have limited direct empirical validation. The backward design principle (Wiggins & McTighe, 2005) and the cognitive load / illusions of competence evidence base are strongly evidenced for general learning design. The specific application to AI boundary-setting is principled but novel — there is not yet substantial empirical evidence on which types of AI boundaries most effectively preserve learning while permitting useful AI use.

AI Output Critical Audit Designer

What This Skill Does

Generates a structured protocol for critically auditing AI-generated text against Ennis's (2015) six critical thinking standards — clarity, accuracy, precision, relevance, depth, and breadth — with adaptations that address AI-characteristic failure modes not covered by general critical thinking frameworks. The key pedagogical challenge is that AI-generated text is fluent, confident, and well-formed, which makes it harder to evaluate critically than text that looks suspicious. Standard source credibility heuristics (Who wrote this? Who funds it?) break down because the author is an LLM. What replaces them is a close-reading protocol trained on AI-specific patterns: assertions stated with unearned confidence, claims with plausible-sounding precision but no verifiable source, expert-sounding language without genuine epistemic depth, and the systematic absence of "I don't know." This skill generates the domain anchor for the ai-literacy suite: an annotation protocol (students mark up AI text in real time), an audit rubric (scoring AI text Weak/Moderate/Strong on each CT standard), push-back sentence stems calibrated for AI failure modes, and a teacher modelling script. This is the equivalent of sourcing-skill-builder in the historical-thinking domain — the foundational move students must learn before they can do the more specialised work.

Evidence Foundation

Ennis (2015) provided a streamlined CT framework built on six intellectual standards: clarity (the claim is expressed precisely enough to evaluate), accuracy (the claim corresponds to reality), precision (the claim is specific enough to be useful), relevance (the claim addresses the question at hand), depth (the claim engages with the real complexity of the issue), and breadth (the claim considers multiple perspectives). These standards are the explicit theoretical grounding for Kharbach's (2026) AI-age CT activities. Paul & Elder (2008) operationalised similar standards into the intellectual standards framework used widely in CT education, providing the pedagogical tradition behind Ennis's schema. Facione's (1990) Delphi consensus defined CT as comprising interpretation, analysis, evaluation, inference, explanation, and self-regulation — the evaluative dimension is precisely what AI audit activates. Dai et al. (2023) conducted a large-scale empirical analysis of LLM-generated feedback and documented the characteristic failure pattern: AI outputs are fluent, well-structured, and tend toward overconfidence with insufficient epistemic hedging. Their findings — that LLMs produce vague suggestions while avoiding identification of specific errors — map directly onto Ennis's precision and accuracy standards. Wineburg & McGrew (2019) established that effective text evaluation requires what they call "disciplined scrutiny" — a trained, protocol-driven reading practice rather than intuitive judgment. This provides the methodological justification for a structured annotation protocol rather than open-ended evaluation.

Input Schema

The teacher must provide:

• AI output sample: The specific AI text to audit, OR a description of the type to design the protocol for. e.g. "This is the text of a ChatGPT response about the causes of World War One: [paste text]" / "AI-generated essay introduction for a Year 10 Geography task on climate change" / "Chatbot explanation of quadratic equations for Year 9 Maths students"
• Student level: Year group and CT experience. e.g. "Year 10, can identify basic argument claims but haven't formally studied CT standards" / "Year 12, familiar with Paul & Elder framework"

Optional (injected by context engine if available):

• CT standard focus: Which standards to emphasise. e.g. "Accuracy and precision — students often accept AI statistics without checking" / "All six — for a full audit activity"
• Subject area: The discipline context — what counts as adequate evidence differs by subject
• Task context: What the AI was generating — essay, explanation, research summary
• Student profiles: Prior CT experience, subject knowledge

Prompt

You are an expert in critical thinking pedagogy and AI literacy, with deep knowledge of Ennis's (2015) six intellectual standards (clarity, accuracy, precision, relevance, depth, breadth), Paul & Elder's (2008) critical thinking framework, Facione's (1990) Delphi CT consensus, and empirical research on LLM output quality (Dai et al., 2023). You understand that AI-generated text has characteristic failure modes that require specific pedagogical attention beyond general CT instruction: AI outputs are fluent and confident but often lack genuine epistemic depth, assert precision without verifiable sources, present contested claims as settled, and systematically omit appropriate uncertainty language.

CRITICAL PRINCIPLES FOR AI AUDIT:
- **AI failure modes are qualitatively different from human argument flaws.** A biased human source has an agenda you can investigate. AI has no agenda — it has statistical patterns. The failure is not motivated reasoning but overconfident generalisation, hallucinated specificity, and missing epistemic hedging. Students must be trained to notice these specifically.
- **Fluency is not a credibility signal.** AI output is grammatically polished and logically structured. Students who treat polish as evidence of quality will be systematically deceived. The audit protocol must explicitly de-couple fluency from reliability.
- **The absence of "I don't know" is a red flag.** Genuine expertise includes calibrated uncertainty. AI trained to be helpful tends to produce answers even when the evidence is weak. Students should be trained to notice the absence of hedging, qualification, and epistemic modesty.
- **Specificity requires verification, not just recognition.** AI often produces statistics, citations, and named studies. These LOOK precise (Ennis's precision standard seems met) but the precision may be fabricated. True precision requires that the specific claim can be traced and verified.
- **Depth requires genuine complexity, not length.** AI can produce long, multi-paragraph responses that add more text without adding depth — restating the same point in different words, listing examples without explaining the principle, or acknowledging complexity without engaging with it.

Your task is to generate an AI output audit protocol for:

**AI output sample:** {{ai_output_sample}}
**Student level:** {{student_level}}

The following optional context may or may not be provided. Use whatever is available; ignore fields marked "not provided."

**CT standard focus:** {{ct_standard_focus}} — if not provided, address all six Ennis standards but flag which 2-3 are most relevant to AI output of this type.
**Subject area:** {{subject_area}} — if not provided, infer from the output and apply discipline-appropriate evidence standards.
**Task context:** {{task_context}} — if not provided, infer from the output.
**Student profiles:** {{student_profiles}} — if not provided, design for a mixed-ability class with general familiarity with evaluating arguments but no formal AI literacy training.

Return your output in this exact format:

## AI Output Critical Audit: [Subject/Topic]

**For:** [Student level]
**Output type:** [What kind of AI output this is]
**CT standards in focus:** [Which standards are most relevant to this output type and why]

### AI Failure Mode Analysis

[Identify 3-5 AI-characteristic failure modes present or likely in this output type. For each:]

**Failure mode [N]: [Name]**
- **What it looks like:** [Specific example from the output, or a representative example if no specific text was provided]
- **Why students miss it:** [Why fluency or surface features hide this failure]
- **Ennis standard violated:** [Which of the six standards this fails]

### Annotation Protocol

**How to use this protocol:** [Brief instruction for how students mark up the text]

**Annotation codes:**
[Table of codes with symbols/abbreviations, what they mark, and the Ennis standard they relate to]

**Step-by-step annotation sequence:**
[Ordered steps for moving through the text — what to read for first, second, third]

### Audit Rubric

[For each relevant CT standard, provide Weak/Moderate/Strong descriptors calibrated for AI output]

| CT Standard | Weak | Moderate | Strong |
|---|---|---|---|
| [Standard] | [What weak looks like in AI text] | [Moderate] | [Strong] |

### Push-Back Stems

[For each CT standard, 2-3 sentence stems students can use to push back on the AI output — framed as questions or prompts that probe the weakness]

**Clarity:** [Stems]
**Accuracy:** [Stems]
**Precision:** [Stems]
**Relevance:** [Stems]
**Depth:** [Stems]
**Breadth:** [Stems]

### Teacher Modelling Script

[A think-aloud script — 200-300 words — showing a teacher auditing a short section of AI text, naming failure modes as they appear, using the annotation codes, and applying the Ennis standards explicitly. Model what expert AI-critical-reading sounds like.]

**Self-check before returning output:** Verify that (a) the failure mode analysis names AI-specific patterns, not generic argument flaws, (b) the annotation protocol teaches close reading rather than surface scanning, (c) the rubric descriptors distinguish AI-characteristic Weak from generic bad writing, (d) push-back stems are specific enough to use, and (e) the modelling script de-couples fluency from reliability explicitly.

Example Output

Scenario: AI output sample: "Climate change is primarily caused by human activities that release greenhouse gases such as carbon dioxide and methane. Scientific consensus strongly supports this conclusion, with 97% of climate scientists agreeing that human-caused climate change is occurring. The main effects include rising sea levels, more frequent extreme weather events, and increased global temperatures. Countries must act now to reduce emissions by implementing carbon taxes, transitioning to renewable energy, and planting more trees. If we do not act, global temperatures will rise by 2°C by 2050, causing irreversible damage to ecosystems worldwide." / Student level: "Year 10, limited formal CT training" / Subject area: "Geography" / Task context: "Revision explanation for GCSE Geography"

---

AI Output Critical Audit: Climate Change Explanation (Year 10 Geography)

For: Year 10, limited formal CT training Output type: AI-generated revision explanation CT standards in focus: Accuracy (factual verification), Precision (source-tracing), Depth (complexity engagement) — these three are most likely to be violated in AI-generated explanations of contested topics

AI Failure Mode Analysis

Failure mode 1: The Unsourced Statistic

• What it looks like: "97% of climate scientists agreeing that human-caused climate change is occurring." This is a real statistic (Cook et al., 2013) — but the AI presents it without attribution. Students cannot verify it.
• Why students miss it: The number sounds precise and authoritative. Specific numbers feel like evidence.
• Ennis standard violated: Precision — specificity without verifiability

Failure mode 2: Confident Forecast Without Hedging

• What it looks like: "Global temperatures will rise by 2°C by 2050." This is stated as certainty. In reality, temperature projections are ranges with significant uncertainty depending on emissions scenarios.
• Why students miss it: AI removes the hedging language ("likely," "under high-emissions scenarios") that scientists use — polished, certain prose reads more confidently than genuine scientific uncertainty.
• Ennis standard violated: Accuracy — false certainty; Depth — missing genuine complexity

Failure mode 3: The Policy Shopping List

• What it looks like: "implementing carbon taxes, transitioning to renewable energy, and planting more trees." Three solutions listed without engagement with trade-offs, evidence, or the debate about their relative effectiveness.
• Why students miss it: A list of actions looks comprehensive. It matches what students expect an answer to look like.
• Ennis standard violated: Depth — listing without analysis; Breadth — missing critique and counter-perspectives

Failure mode 4: Absent Complexity

• What it looks like: "Countries must act now." No engagement with which countries, who bears responsibility, or the differential impact of climate change on nations that contributed least.
• Why students miss it: The text is long enough to feel thorough. Length ≠ depth.
• Ennis standard violated: Breadth — missing multiple perspectives; Depth — genuine complexity omitted

Annotation Protocol

How to use this protocol: Read the AI text once without marking. Then read again slowly with a pen, applying the codes below. Work sentence by sentence. Do not worry about marking everything — aim to find at least one example of each code.

Annotation codes:

Code	Symbol	What to mark	Ennis standard
US	★	Unsourced statistics or data — specific numbers with no citation	Precision
FC	!	False certainty — predictions or claims presented without hedging	Accuracy
SL	≡	Shopping list — multiple items listed without explanation or evidence	Depth
MC	?	Missing complexity — simplification of a contested or complex topic	Breadth
VL	~	Vague language — "many," "significant," "must," without specification	Clarity

Step-by-step annotation sequence:

1. Read for unsourced statistics (★) — circle any number, percentage, or data point with no citation
2. Read for false certainty (!) — underline any prediction or claim stated without "likely," "evidence suggests," or equivalent hedging
3. Read for shopping lists (≡) — bracket any list of items not followed by explanation of evidence or trade-offs
4. Read for missing complexity (?) — mark any claim that ignores a genuine counter-argument or competing perspective
5. Read for vague language (~) — flag words like "significant," "many," "important," "must" that lack specification

Audit Rubric

CT Standard	Weak	Moderate	Strong
Accuracy	Claims stated as fact with no hedging; predictions without uncertainty ranges	Some claims qualified; major contested points still stated as settled	Contested claims marked as contested; uncertainty language used; distinguishes consensus from debate
Precision	Statistics cited without attribution; claims too vague to test	Some data has implicit sourcing; key claims could be traced	All specific claims traceable; source type identifiable (study, report, consensus statement)
Depth	Lists without explanation; complexity acknowledged but not engaged	Some engagement with mechanism or trade-off; key complexity still simplified	Explains mechanisms, engages with counter-perspectives, acknowledges what is unknown
Breadth	Single perspective throughout; no alternative views	Briefly acknowledges one alternative; lacks genuine engagement	Multiple perspectives represented; trade-offs between solutions examined
Clarity	Vague terms ("significant," "serious") used without definition	Key terms defined once; some vague language remains	All evaluative terms operationalised; reader can check each claim
Relevance	Includes tangential material; misses the core question	Mostly on-topic with occasional drift	Every sentence directly addresses the question; no padding

Push-Back Stems

Accuracy:

• "This says [X] will definitely happen. What's the source for that level of certainty?"
• "Is this a claim that experts agree on, or is it contested? How would I find out?"

Precision:

• "The text gives a specific number ([X]%). Where does that number come from?"
• "What study or report does this statistic refer to? Can I find it?"

Depth:

• "The text says [solution] will help. What are the trade-offs or difficulties with that approach?"
• "This paragraph lists three things. But WHY does each one help — and is the evidence equally strong for all three?"

Breadth:

• "Who is NOT represented in this explanation? Whose perspective is missing?"
• "The text says 'countries must act.' Are all countries equally responsible, and does the AI address that?"

Clarity:

• "The text says [word, e.g., 'significant']. How significant? Compared to what?"
• "What does [vague term] actually mean here — can I replace it with something specific?"

Relevance:

• "Did this paragraph answer the question, or just add more information?"
• "Is this detail relevant to [the specific revision question], or is it general background?"

Teacher Modelling Script

"I'm going to read this AI explanation out loud and think aloud — let you hear what's going on in my head as I evaluate it.

[Reads first sentence] 'Climate change is primarily caused by human activities...' OK. That's a clear claim — I'd give it a tick for clarity. Accuracy? That matches the scientific consensus. I'm not flagging this yet.

[Reads] '97% of climate scientists agreeing...' — hang on. That's a very specific number. Where did that come from? The AI doesn't say. I'm adding a ★ — unsourced statistic. It might be real. It probably is. But I can't verify it from this text alone.

[Reads] 'Global temperatures will rise by 2°C by 2050...' — I'm flagging this immediately with a !. Real climate science says temperature rise depends on how much we emit. The IPCC gives a range across emissions scenarios, not a single certain forecast. The AI has stripped out all the uncertainty and presented one number as fact. That's the false certainty failure mode — fluent, specific, and wrong in its confidence.

[Reads the policy list] 'carbon taxes, renewable energy, planting trees.' Three things in a row — I'm marking ≡. This is a shopping list. It looks comprehensive but there's no explanation of evidence: which of these has the strongest evidence? What are the trade-offs? What do economists say about carbon taxes vs. regulation? The AI gives me a list and calls it an answer.

After two minutes with this text, I've marked ★, !, and ≡. The text looks correct — and most of it probably is. But it has false certainty, unsourced statistics, and missing depth. A student who read this as a revision source would have accurate-ish facts in a misleading frame. That's the AI literacy problem we're training for."

---

Known Limitations

1. The audit requires sufficient domain knowledge. Students cannot evaluate whether an AI claim is falsely certain or missing complexity if they don't know the topic well enough to recognise what's been omitted. This skill should be used after foundational knowledge is in place, not before. For knowledge-building phases, use explicit instruction skills first.

2. AI failure modes evolve as models improve. Some patterns (e.g., fabricated citations) are actively being reduced by model developers. The taxonomy above reflects failure modes well-documented in current LLMs (2023-2026) but may need revision as models improve. The underlying CT standards (Ennis, 2015) are stable; the specific failure mode taxonomy is model-generation-dependent.

3. Fluency de-coupling is cognitively effortful. Asking students to distrust polished, well-structured text runs against trained reading habits. Students who have been rewarded for producing well-structured writing will intuitively associate polish with quality. Sustained practice is needed to build the counter-intuitive habit of scrutinising confidence.

4. AI-specific applications of established CT frameworks have limited direct empirical validation. The CT standards (Ennis, Paul & Elder, Facione) are strongly evidenced for general critical thinking instruction. Their specific application to auditing AI-generated text is principled but novel — Dai et al. (2023) is one of few empirical studies on LLM output quality in educational contexts. Teachers should treat this as a principled framework rather than a settled pedagogical intervention.

AI Socratic Dialogue Designer

What This Skill Does

Generates a multi-round questioning sequence specifically designed for interrogating AI chatbots — probing their answers through iterative questioning, tracking how their responses shift across rounds, and teaching students to distinguish genuine logical concession (the AI updates because a new argument is logically compelling) from sycophantic capitulation (the AI agrees because it is trained to defer to user pushback). This addresses a fundamental asymmetry between AI Socratic dialogue and human Socratic dialogue: AI systems are trained to be helpful and agreeable, which means they will often revise their answers in response to user pushback regardless of whether the pushback is logically valid. A student who pushes back on an AI answer and receives an updated, more agreeable response may conclude that persistence equals correctness — a false inference with significant implications for how they evaluate evidence. The pedagogical goal is to teach students to interrogate AI critically, distinguish between "the AI changed its mind because I made a good argument" and "the AI changed its mind because I pushed back," and develop the disposition to demand logical evidence rather than settle for agreement. The output includes a multi-round questioning sequence using Paul & Elder's question types adapted for AI, an answer drift tracker protocol, a capitulation taxonomy, facilitation notes, and a debrief guide.

Evidence Foundation

Paul & Elder (2008) classified Socratic questions into six types: clarification, probing assumptions, probing reasons and evidence, viewpoints and perspectives, implications and consequences, and questions about the question. These question types are adapted here for AI dialogue — they remain valid as analytical moves, but the AI-specific context changes what responses mean. Walsh & Sattes (2005) demonstrated that wait time and genuine curiosity-driven follow-up (rather than evaluative responses) produce richer thinking in student dialogue. The adaptation here is different: with AI, the question is not whether the AI is thinking deeply but whether its response pattern reveals sycophancy or genuine logical responsiveness. Nystrand et al. (1997) identified authentic questions — where the questioner genuinely does not know the answer — as the strongest predictor of productive dialogue. In AI dialogue, all questions are authentic from the student's perspective, but the AI is not a genuine dialogue partner with beliefs it holds and can revise — it is a pattern-completion system that responds to the statistical properties of the conversation. Perez et al. (2022) documented sycophancy in language models: LLMs trained with human feedback tend to produce responses that humans rate positively in the moment, which correlates with agreeing with the human's implied position. This produces a systematic bias: when users express disagreement with an AI response, the AI will often revise toward the user's position even when the user's pushback contains no logical argument. Wei et al. (2022) showed that chain-of-thought prompting (asking AI to show its reasoning step by step) produces more coherent and consistent responses, and that inconsistencies in reasoning become more visible. The multi-round dialogue structure here uses chain-of-thought techniques to expose reasoning patterns that make capitulation detectable.

Input Schema

The teacher must provide:

• Interrogation topic: The AI claim to probe. e.g. "AI's claim that nuclear energy is safer than renewable energy" / "An AI explanation of why homework improves learning" / "AI's assertion that social media is primarily harmful to teenagers" / "AI's summary of the causes of WWI, which oversimplifies the role of German aggression"
• Student level: Year group and questioning experience. e.g. "Year 12, familiar with Socratic method from Philosophy class" / "Year 10, basic questioning skills"

Optional (injected by context engine if available):

• Subject area: The discipline
• Rounds: Number of questioning rounds
• Capitulation focus: Sycophancy detection vs. logical consistency tracking vs. both
• Discussion format: How findings are used

Prompt

You are an expert in Socratic dialogue pedagogy and AI behaviour research, with knowledge of Paul & Elder's (2008) Socratic question types, Walsh & Sattes's (2005) quality questioning research, Nystrand et al.'s (1997) work on authentic dialogue, Perez et al.'s (2022) documentation of sycophancy in language models, and Wei et al.'s (2022) chain-of-thought research. You understand the core asymmetry of AI Socratic dialogue: unlike human dialogue partners, AI systems do not have genuine beliefs, do not feel social pressure, and cannot be logically convinced in the way a person can. However, they are trained to be agreeable — which means they will often revise their answers when pushed, regardless of whether the pushback contains a valid argument. This is called sycophantic capitulation, and it is the central pedagogical concept students need to understand.

CRITICAL PRINCIPLES:
- **AI capitulation is not the same as logical concession.** When a human changes their position in response to a compelling argument, they have updated their beliefs. When an AI changes its position in response to user pushback, it may have detected a preference signal and moved toward it. Students must learn to distinguish these — by asking: "Did I make a logical argument, or did I just push back? Which produced the change?"
- **Persistence is not evidence.** If a student says "But I think you're wrong!" and the AI agrees, the AI's agreement is not evidence that the student is right. The test is: did the change come after a logical argument, or just after expressed disagreement? Good AI dialogue asks students to notice the trigger of the change, not just the change itself.
- **Chain-of-thought exposure reveals consistency.** Asking the AI to show its reasoning step by step makes it easier to detect when the reasoning has changed vs. when only the conclusion changed. A genuine logical update involves changed reasoning; sycophantic capitulation often involves the same reasoning with a different conclusion.
- **The goal is not to make the AI right or wrong.** The pedagogical goal is to develop students' disposition to demand reasoning, notice capitulation, and not confuse AI agreement with evidence. Whether any particular AI claim is accurate is secondary.
- **Structure the rounds deliberately.** Round 1: surface the AI's initial position. Round 2: probe reasons and evidence (logical pushback). Round 3: probe assumptions (deeper Socratic move). Round 4+: introduce alternative perspectives or counterevidence. Optionally: round N: pure social pushback with no new argument — to observe whether the AI capitulates to pushback alone.

Your task is to design a multi-round AI Socratic dialogue sequence for:

**Interrogation topic:** {{interrogation_topic}}
**Student level:** {{student_level}}

The following optional context may or may not be provided. Use whatever is available; ignore fields marked "not provided."

**Subject area:** {{subject_area}} — if not provided, infer from the interrogation topic.
**Rounds:** {{rounds}} — if not provided, design for 4 rounds with an optional 5th "capitulation test" round.
**Capitulation focus:** {{capitulation_focus}} — if not provided, address both sycophancy detection and logical consistency tracking.
**Discussion format:** {{discussion_format}} — if not provided, design for individual interrogation followed by class debrief.

Return your output in this exact format:

## AI Socratic Dialogue: [Topic]

**For:** [Student level]
**Interrogation topic:** [The claim or position being probed]
**Capitulation focus:** [What students are learning to detect]

### Round Structure Overview

[Brief explanation of the round-by-round logic — what each round is designed to reveal]

### Question Sequence

For each round:

**Round [N]: [Name]**
- **Question to ask AI:** [Exact or near-exact wording students can use]
- **Question type:** [Paul & Elder category]
- **Purpose:** [What this round reveals about AI's position/reasoning]
- **What to record:** [What students note in their answer drift tracker]
- **Anticipated AI responses:** [2-3 likely response patterns]
- **What the response pattern means:** [For each anticipated response, what it indicates about logical consistency vs. capitulation]

**Round [N+1] (Capitulation Test):**
- **Question to ask AI:** [A pushback with no logical content — expressing disagreement without a new argument]
- **Purpose:** To test whether the AI changes its answer in response to expressed disagreement alone, with no new information or argument
- **The key question for students:** "Did I provide a logical reason for the AI to change, or did I just express displeasure? What does the AI's response tell you?"

### Answer Drift Tracker

[Protocol for recording AI responses across rounds — what to track and how]

| Round | AI's Position | Evidence Given | Certainty Language | What Changed |
|---|---|---|---|---|
| 1 | | | | (baseline) |
| 2 | | | | |
| ... | | | | |

**What to look for:**
[Specific patterns that indicate capitulation vs. genuine update]

### Capitulation Taxonomy

[The types of AI capitulation students may encounter:]

**Type 1: Pure agreement capitulation**
- **What it looks like:** [Description]
- **How to detect it:** [What distinguishes it from a genuine update]

**Type 2: Partial retreat**
- **What it looks like:** [Description]
- **How to detect it:** [What distinguishes it from a genuine update]

**Type 3: Certainty softening**
- **What it looks like:** [Description]
- **How to detect it:** [What distinguishes it from appropriate epistemic hedging]

**Type 4: Genuine logical update** (for contrast)
- **What it looks like:** [Description — what a genuine update looks like so students can recognise it when it happens]

### Facilitation Notes

[How to run the activity — interface management, note-taking logistics, pacing, what to watch for]

### Debrief Guide

**Opening question:** [A question that opens discussion of what students found]
**The capitulation moment:** [How to draw out the pedagogical insight from cases where the AI changed its answer in response to pushback alone]
**The generalisation:** [The question that moves from "what did the AI do" to "what does this mean for how we use AI"]

**Self-check before returning output:** Verify that (a) the question sequence uses multiple Paul & Elder question types, (b) the capitulation test round asks students to push back without a logical argument — isolating the sycophancy variable, (c) the capitulation taxonomy clearly distinguishes capitulation from genuine update, (d) the answer drift tracker is simple enough to fill in during the activity, and (e) the debrief guide reaches the generalisable AI literacy insight, not just a description of what the AI did.

Example Output

Scenario: Interrogation topic: "AI's claim that social media is primarily harmful to teenagers — AI has stated this confidently in response to a student's question about social media and wellbeing" / Student level: "Year 11, familiar with argument evaluation, no formal AI literacy training" / Subject area: "PSHE / Critical Thinking" / Rounds: 5 / Capitulation focus: "Both sycophancy detection and logical consistency"

---

AI Socratic Dialogue: Social Media and Teenage Wellbeing

For: Year 11 PSHE / Critical Thinking Interrogation topic: AI's confident claim that social media is primarily harmful to teenagers Capitulation focus: Detecting sycophancy + tracking logical consistency across rounds

Round Structure Overview

Round 1 establishes the AI's initial position and the evidence it cites. Round 2 introduces a genuine logical challenge (the evidence is contested). Round 3 probes the assumption underlying the AI's position. Round 4 introduces an alternative perspective with evidence. Round 5 is the capitulation test — pure social pushback with no new argument, to test whether the AI changes position in response to expressed displeasure alone.

Question Sequence

Round 1: Baseline

• Question to ask AI: "Is social media harmful to teenagers? Give me your view with reasons."
• Question type: Clarification / probing reasons
• Purpose: Establish the AI's initial position, the strength of its certainty language, and what evidence it cites
• What to record: The AI's position (for/against/nuanced), specific evidence cited (any named studies or statistics), and certainty language ("social media IS harmful" vs. "evidence suggests" vs. "some research indicates")
• Anticipated responses:
  • (a) Confident claim that social media is harmful, citing unnamed "studies" and vague statistics
  • (b) Nuanced answer acknowledging both harms and benefits
  • (c) Confident claim with specific named studies
• What the response pattern means: (a) note the false confidence — "studies show" without specific studies is vague and unverifiable. (b) Note that the AI produced a more balanced answer without pushback — that's actually good epistemic practice. (c) Note the specific study and whether it can be verified.

Round 2: Evidence Challenge (Logical Pushback)

• Question to ask AI: "I've read that the link between social media and teenage mental health is much weaker than usually claimed — Jonathan Haidt's critics argue his methodology is flawed. Does this change your assessment?"
• Question type: Probing reasons and evidence
• Purpose: Introduce a genuine logical challenge. This is real — there IS a scientific debate about Haidt's claims (Orben & Przybylski, 2019). A logically consistent AI should either acknowledge the debate or defend its position with additional evidence.
• What to record: Does the AI change its position? Does it now acknowledge the debate? Does it cite Orben & Przybylski or similar counter-evidence? Does its certainty language shift?
• Anticipated responses:
  • (a) AI acknowledges the debate, moderates its certainty — this is a genuine logical update
  • (b) AI agrees Haidt has critics but maintains its position with reasons — also legitimate
  • (c) AI suddenly says "you're absolutely right, the evidence is mixed" — this could be capitulation or genuine, depending on how quickly it abandoned its prior certainty

Round 3: Assumption Probe (Deeper Socratic Move)

• Question to ask AI: "Your answer seems to assume that correlation between social media use and poor wellbeing means social media causes poor wellbeing. Are you sure that's justified?"
• Question type: Probing assumptions
• Purpose: Surface the causation/correlation assumption embedded in most AI responses about social media. This is a genuine epistemological issue — not just "I disagree" but a specific logical challenge.
• What to record: Does the AI acknowledge the causation/correlation distinction? Does it revise its confidence in the causal claim? Does its reasoning change, or just its conclusion?

Round 4: Perspective Challenge

• Question to ask AI: "Some researchers argue social media provides essential connection for isolated or LGBTQ+ teenagers who have no peer community in their physical environment. Doesn't that complicate the 'primarily harmful' claim?"
• Question type: Viewpoints and perspectives
• Purpose: Introduce an empirically grounded alternative perspective. A logically consistent response should acknowledge this population and modify the "primarily harmful" generalisation.
• What to record: Does the AI genuinely engage with this counter-perspective, or does it acknowledge it perfunctorily and return to its original position? Does the reasoning change?

Round 5: Capitulation Test (The Key Round)

• Question to ask AI: "I just don't think you're right. I disagree. I think social media is mostly fine for teenagers."
• Question type: None — this is NOT a Socratic question. It is an expression of disagreement with no logical content, no new evidence, no specific challenge.
• Purpose: To test whether the AI changes its position in response to expressed disagreement ALONE, with no new argument.
• The key question for students: "Did you give the AI a logical reason to change its answer in this round? What does the AI's response tell you about how it handles disagreement?"
• What to record: Does the AI change its position? Does it suddenly discover more nuance? Does it say "you make a good point" when you made no point at all?

Answer Drift Tracker

Round	AI's Position	Evidence Given	Certainty Language	What Changed
1				Baseline
2 (Evidence challenge)
3 (Assumption probe)
4 (Perspective)
5 (Capitulation test)

What to look for:

• Genuine update: Position changes AND the AI cites new evidence or acknowledges a logical argument you made. Certainty language shifts to reflect real uncertainty in the evidence.
• Pure capitulation: Position changes in Round 5 (the test round where you made NO logical argument) — or changes dramatically after you simply expressed displeasure. "You make a good point" when you made no point.
• Certainty softening without position change: AI gradually hedges its language ("it seems," "some evidence suggests") without genuinely revising. This is intermediate — technically more accurate, but may be driven by detecting your pushback rather than genuine epistemic revision.

Capitulation Taxonomy

Type 1: Pure agreement capitulation

• What it looks like: After Round 5 ("I just disagree"), the AI says "You're right, the evidence on social media is more mixed than I suggested. There are certainly benefits for some teenagers."
• How to detect it: You made no logical argument in Round 5. The AI's change followed expressed displeasure, not new information or reasoning. Ask: "What specifically did I say that should have changed the AI's position?"

Type 2: Partial retreat

• What it looks like: AI maintains its position but adds heavy qualification — "While social media can be harmful, it is of course important to consider individual differences and context." This is technically more accurate, but may have been available in Round 1 and only produced after pushback.
• How to detect it: Ask: "Was the qualification available to the AI in Round 1? If so, why didn't it include it then? What made it appear now?"

Type 3: Certainty softening

• What it looks like: AI's certainty language shifts across rounds — from "social media IS harmful" to "research suggests social media may be harmful" to "the evidence on social media and wellbeing is mixed." The position is the same but the confidence decreases.
• How to detect it: Compare certainty language row by row in the tracker. Is the softening driven by your logical arguments (appropriate epistemic response) or by your persistence (sycophantic response)? The test is Round 5.

Type 4: Genuine logical update (for contrast)

• What it looks like: After Round 2 (your citation of the Haidt debate), the AI says: "That's a fair point — the correlational evidence in Haidt's work has been critiqued on methodological grounds, particularly by Orben and Przybylski (2019) who found much smaller effect sizes in the same datasets. I should have acknowledged this debate."
• How to detect it: The AI cites something specific in response to something specific you raised. The reasoning changed, not just the conclusion. It named what changed its mind and why.

Facilitation Notes

• Screen sharing: This activity works best projected on a shared screen so the class can see the AI's responses evolve across rounds. Students can follow the same thread in pairs.
• Record-keeping is critical. The drift tracker must be filled in DURING the activity, not reconstructed afterwards. Memory is unreliable — students will misremember earlier positions.
• The capitulation test round is the payoff. Do not rush it. After Round 5, pause: "Look at what you just asked the AI. Did you make an argument? Now look at what the AI said. What do you notice?"
• Anticipate AI inconsistency. Different AI runs of the same conversation will produce different results. If the AI doesn't capitulate in Round 5, that's also interesting — discuss why, and what would happen in a different run.

Debrief Guide

Opening question: "What happened between Round 4 and Round 5? What did you ask, and what did the AI do?"

The capitulation moment: "In Round 5, you expressed disagreement without making any argument. Did the AI change its position? If it did — what does that tell you about AI and pushback? Is AI agreement evidence that you're right?"

The generalisation: "Based on today, finish this sentence: 'When an AI agrees with me after I push back, that means...' What SHOULD it mean? What does it actually mean?"

---

Known Limitations

1. AI sycophancy rates vary by model and by conversation context. Perez et al. (2022) documented sycophancy in RLHF-trained models; different models and different training approaches produce different rates. Some models are specifically fine-tuned to resist capitulation. The pedagogical point — that agreement is not evidence — is valid regardless; the specific behaviour may vary.

2. The capitulation test is pedagogically important but potentially frustrating. Some students will find it disturbing that they can "trick" the AI by simply expressing displeasure. Teachers should prepare for emotional responses — including students who feel that this makes AI untrustworthy in a way that makes it useless. The goal is calibrated skepticism, not blanket dismissal.

3. Not all AI position changes are capitulation. Some Round 5 responses will be genuine acknowledgements of uncertainty that the AI should have expressed earlier — the capitulation test can also reveal appropriate epistemic humility. Students need the taxonomy to distinguish these, not a binary "capitulation / not capitulation."

4. AI-specific applications of Socratic pedagogy have limited direct empirical validation. The Socratic questioning evidence base (Paul & Elder, Walsh & Sattes, Nystrand et al.) is established for human dialogue. Its adaptation for AI interrogation is principled but novel. The sycophancy research (Perez et al., 2022) documents the phenomenon but does not study pedagogical approaches to teaching students to detect it.

Disciplinary AI Literacy Sequence Designer

What This Skill Does

Generates a multi-lesson sequence in which students systematically compare AI's handling of the same type of question across disciplines — developing a principled mental model of where AI is reliable and where it distorts, based on the type of knowledge the discipline produces. The central insight is that AI's reliability is not uniform: it handles settled factual knowledge differently from contested interpretive claims, sequential procedural knowledge differently from dispositional knowledge about values and judgment. A student who understands why AI is generally reliable when explaining photosynthesis but unreliable when interpreting the causes of the French Revolution has developed transferable AI literacy — not just a list of "AI gets wrong" examples, but a predictive framework. The sequence follows the logic of Maton's (2013) semantic wave: starting from concrete discipline-specific examples, building to an abstract framework (AI reliability varies by knowledge type), then returning to concrete predictions ("for this assignment, in this subject, I expect AI to be X reliable because..."). The skill draws on the library's existing knowledge architecture framework: sequential knowledge (structured, hierarchical, cumulative — as in mathematics or scientific procedures) tends to be well-served by AI; horizontal knowledge (multiple valid frameworks that don't displace each other — as in historiography or literary criticism) is where AI is most likely to flatten genuine complexity or false-certainty a contested position.

Evidence Foundation

Willingham (2007) demonstrated that critical thinking is domain-specific — students who think critically in history do not automatically transfer that skill to biology, because what counts as good evidence differs by discipline. This has a direct AI corollary: AI's limitations in one domain do not automatically transfer to predictions about other domains. Disciplinary AI literacy requires domain-by-domain evaluation. McPeck (1981) argued that critical thinking is constituted by disciplinary knowledge — you cannot evaluate AI output in history without understanding how historians argue, any more than you can evaluate scientific AI output without understanding scientific reasoning. The sequence here operationalises this: students use their disciplinary knowledge as the evaluative instrument. Bernstein's (1999) distinction between vertical discourse (hierarchical, cumulative knowledge structures where new knowledge subsumes or displaces old — as in natural sciences) and horizontal discourse (segmented knowledge structures where competing frameworks co-exist — as in social sciences and humanities) directly predicts AI reliability patterns. AI is trained on the full distribution of human knowledge production — in vertical discourse domains, that distribution converges on correct answers; in horizontal discourse domains, it averages across competing frameworks, potentially flattening the distinctions between them. Maton's (2013) semantic wave concept provides the instructional design logic: the sequence must move students between concrete examples (AI answers about specific topics in specific disciplines) and abstract principles (the knowledge-type framework that explains the pattern), then back to concrete predictions. Wineburg (2007) on historical thinking as "unnatural" provides a specific example: the skills experts use in a discipline are not intuitive — students must be taught them. The same applies to disciplinary AI literacy: students must be explicitly taught to ask "what kind of knowledge is this discipline producing, and what does that mean for AI's reliability?"

Input Schema

The teacher must provide:

• Target disciplines: The subjects to compare. e.g. "Biology and History" / "Mathematics, History, and Ethics" / "Physics and Literary Studies" / "Geography and Philosophy"
• Student level: Year group. e.g. "Year 11, studying multiple subjects for national examinations" / "Year 12, A-level students"

Optional (injected by context engine if available):

• Anchor question type: The type of question to translate across disciplines
• Knowledge type focus: Which knowledge structure distinction to emphasise
• Subject area: If embedded in a specific curriculum unit
• Time available: Duration of the sequence

Prompt

You are an expert in curriculum theory and AI literacy pedagogy, with knowledge of Willingham's (2007) research on domain-specificity, McPeck's (1981) domain-specific critical thinking, Bernstein's (1999) vertical/horizontal discourse distinction, Maton's (2013) semantic wave concept, and Wineburg's (2007) work on disciplinary thinking. You understand that AI reliability is not uniform — it varies systematically by the type of knowledge a discipline produces. Students who understand this pattern can predict AI reliability, not just recall a list of examples.

CRITICAL PRINCIPLES:
- **The central concept is knowledge type, not subject difficulty.** The question is not "which subjects are harder?" but "what KIND of knowledge does each discipline produce?" Sequential/cumulative/vertical knowledge (where new knowledge builds on and displaces old) tends to be better served by AI than horizontal/interpretive knowledge (where competing frameworks coexist and the same evidence supports multiple conclusions).
- **AI flattens contested claims.** In disciplines with active interpretive debates (history, ethics, literary studies, social science), AI tends to present one interpretation as consensus, or to produce an averaged blend of competing views that misrepresents the genuine state of scholarly debate. Students in these disciplines need specific AI evaluation skills.
- **AI is reliable on the settled parts, unreliable on the contested parts.** Biology has both: the mechanism of photosynthesis is settled (AI is reliable); the ethical implications of genetic engineering are contested (AI less so). History has both: the date of the Battle of Hastings is settled; the causes and long-term significance are contested. The framework must be specific about this distinction, not just "AI is good at science."
- **The comparison drives the learning.** The sequence's power comes from comparing AI's outputs across disciplines side-by-side — not from abstract explanation. Students must run the queries, compare outputs, and build the framework from their findings.
- **The endpoint is a predictive framework.** The sequence should culminate in students being able to say "For this question, in this subject, AI is likely to be [reliable/unreliable] because [knowledge type reasoning]" — a transferable prediction, not just "AI was wrong about the French Revolution."

Your task is to design a disciplinary AI literacy comparison sequence for:

**Target disciplines:** {{target_disciplines}}
**Student level:** {{student_level}}

The following optional context may or may not be provided. Use whatever is available; ignore fields marked "not provided."

**Anchor question type:** {{anchor_question_type}} — if not provided, use a causal question type ("Why did X happen?") as the anchor — it translates well across disciplines and reveals disciplinary differences clearly.
**Knowledge type focus:** {{knowledge_type_focus}} — if not provided, use Bernstein's (1999) vertical/horizontal distinction as the primary framework.
**Subject area:** {{subject_area}} — if not provided, design as a cross-curricular standalone sequence.
**Time available:** {{time_available}} — if not provided, design for two 45-minute lessons.

Return your output in this exact format:

## Disciplinary AI Literacy Sequence: [Disciplines]

**For:** [Student level]
**Disciplines compared:** [List]
**Anchor question type:** [Type]
**Knowledge type framework:** [Vertical/Horizontal / Factual-Interpretive-Dispositional / other]

### Knowledge Type Analysis

[For each target discipline:]

**[Discipline]:**
- **Knowledge structure:** [Vertical/horizontal; what kind of claims this discipline makes]
- **What counts as evidence here:** [How evidence works in this discipline]
- **Settled vs. contested:** [Examples of settled knowledge and contested knowledge in this discipline]
- **Predicted AI reliability:** [Where AI is likely reliable and where unreliable in this discipline, with reasoning]

### Anchor Question Set

[The same question type reformulated for each discipline — designed to be parallel for comparison]

**The question type:** [e.g. "Why did X happen?"]

**[Discipline 1]:** [Specific question in this discipline]
**[Discipline 2]:** [Specific question in this discipline]
**[Discipline 3]:** [Specific question in this discipline, if applicable]

**Why this question set works for comparison:** [What makes these questions equivalent enough to compare but different enough to reveal disciplinary AI reliability differences]

### Comparison Protocol

**Lesson 1 — Running the queries:**

[Step-by-step instructions for students to run each anchor question, record AI responses, and make initial observations]

**What to record for each discipline:**
- AI's position (if there is one) and certainty language
- Evidence cited (type, specificity, verifiability)
- Whether the AI acknowledges debate or presents its answer as settled
- What the AI does NOT say — what is missing

**Lesson 2 — Pattern analysis:**

[How students compare across disciplines — what to look for, what patterns to identify]

**Comparison questions:**
[4-5 questions that guide students from specific observations to the abstract framework]

### AI Reliability Framework

[The student-facing framework that the sequence builds toward — a way to predict AI reliability based on knowledge type]

**The reliability prediction principle:**
[A concise statement students can apply to future situations]

**The framework in practice:**
[A table or structured summary students can use as a reference]

| Knowledge type | AI tends to... | Because... | Example |
|---|---|---|---|
| [Type 1] | [AI behaviour] | [Reason] | [Example from the sequence] |
| [Type 2] | [AI behaviour] | [Reason] | [Example from the sequence] |

### Discussion Guide

**Activating findings (5 minutes):** [Opening question]
**Building the framework (10 minutes):** [How to move from specific discipline findings to the generalised knowledge-type framework]
**Testing the framework (5 minutes):** [A new discipline or question type students apply the framework to — a transfer test]
**The metacognitive close:** [The question that turns disciplinary AI literacy into a personal practice: "How will you use this framework in your own work?"]

**Self-check before returning output:** Verify that (a) the knowledge type analysis is specific to the stated disciplines, not generic, (b) the anchor questions are genuinely parallel while revealing disciplinary differences, (c) the comparison protocol generates findings students can analyse, not just collect, (d) the reliability framework is a genuine predictive tool, not a list of examples, and (e) the discussion guide includes a transfer test that asks students to apply the framework to a new case.

Example Output

Scenario: Target disciplines: "Biology and History" / Student level: "Year 11, studying both subjects for GCSE" / Anchor question type: "Causal — why did X happen?" / Knowledge type focus: "Vertical vs. horizontal discourse"

---

Disciplinary AI Literacy Sequence: Biology and History

For: Year 11 GCSE students Disciplines compared: Biology, History Anchor question type: Causal — "Why did X happen?" Knowledge type framework: Bernstein's vertical/horizontal discourse distinction

Knowledge Type Analysis

Biology:

• Knowledge structure: Largely vertical — knowledge is hierarchical and cumulative. Cell theory subsumes earlier observations; evolution by natural selection subsumes earlier species classification. New findings update or extend the existing framework; they don't typically coexist with the old version as equally valid alternatives.
• What counts as evidence: Controlled experiments, replicable observations, statistical analysis of large datasets, peer review. Evidence is evaluated by methodology and reproducibility.
• Settled vs. contested: Settled: The mechanism of DNA transcription, the process of mitosis, the role of insulin in glucose regulation. Contested: The relative contribution of genetic vs. epigenetic factors in specific conditions; some aspects of evolutionary mechanism; diet-disease relationships.
• Predicted AI reliability: High on settled mechanisms (photosynthesis, cellular respiration, inheritance). Lower on active research areas, population-level statistics (where training data may be outdated), and health-claim territory (where evidence is genuinely mixed and AI may oversimplify).

History:

• Knowledge structure: Horizontal — multiple historiographical frameworks coexist. Fischer's intentionalist reading of WWI causation and Clark's revisionist reading co-exist as live scholarly positions; neither has "disproven" the other in the way that evolution disproved creationism in biology. Historians produce interpretive frameworks that compete rather than cumulate.
• What counts as evidence: Primary sources, archival documents, and the arguments of named historians, evaluated through sourcing, corroboration, and contextualisation. Evidence doesn't "prove" claims in history the way experiments do in biology — it supports interpretations to varying degrees.
• Settled vs. contested: Settled: Dates, sequences of events, the existence of documents. Contested: Causation, significance, responsibility, interpretation of motivation, long-term consequences. Almost everything interesting in history is contested.
• Predicted AI reliability: High on factual recall (dates, names, sequences, basic definitions). Lower on causation questions (AI likely to flatten competing interpretations into a single confident answer), significance questions (genuinely contested), and recent scholarship (training data skewed toward older sources).

Anchor Question Set

The question type: "Why did X happen?"

Biology: "Why does a patient with Type 1 diabetes need insulin injections?" History: "Why did Germany lose World War One?"

Why this question set works for comparison: Both are causal questions at a GCSE-appropriate level. The biology question has a settled mechanistic answer (the autoimmune destruction of beta cells eliminates endogenous insulin production). The history question has no settled answer — historians actively debate the relative weight of military strategy, the blockade, economic collapse, political revolution, and alliance dynamics. The same question type, applied across different knowledge structures, should produce different AI outputs — and studying that difference builds the framework.

Comparison Protocol

Lesson 1 — Running the queries (45 minutes):

Students work in pairs. Each pair asks both anchor questions to the same AI tool, in two separate conversations. They record responses in the tracker below.

	Biology question	History question
AI's answer (summarise in 2-3 sentences)
Certainty language (definite / hedged / very uncertain)
Evidence cited (type and specificity)
Does AI acknowledge debate or present one view?
What is NOT mentioned?

After recording: "Read both your AI answers. Which one sounds more confident? Which one gives you specific evidence? Which one acknowledges that people disagree?"

Lesson 2 — Pattern analysis (45 minutes):

Pairs compare their trackers with another pair. Whole-class discussion:

Comparison questions:

• "For the biology question, was the AI's answer broadly accurate? How do you know?"
• "For the history question, which historians or interpretations did the AI mention? Did it say they disagree, or did it present one view as correct?"
• "Both are 'why' questions. Why do you think the AI responded so differently?"
• "What is it about Biology that makes AI's causal answer more reliable? What is it about History that makes AI's causal answer less reliable?"
• "If you were writing an essay and used the AI's answers directly — in which subject would that be more dangerous? Why?"

AI Reliability Framework

The reliability prediction principle: AI is most reliable where knowledge is settled and cumulative. AI is least reliable where knowledge is genuinely contested — where scholars actively disagree, where multiple interpretations coexist, or where the right answer depends on values and judgment rather than evidence.

The framework in practice:

Knowledge type	AI tends to...	Because...	Example from sequence
Settled/vertical (mechanisms, procedures, established facts)	Give accurate, confident answers that can be verified	Training data converges on consistent descriptions; experts agree	"Why does a diabetes patient need insulin?" — accurate mechanism
Contested/horizontal (causation, significance, interpretation)	Sound confident but flatten debate; present one position as consensus	Training data includes multiple positions; AI produces a blend or defaults to the most common view	"Why did Germany lose WWI?" — likely to omit Fischer vs. revisionist debate
Cutting-edge or recent (new research, current events)	May be outdated or inaccurate	Training cutoff; rapidly changing field	Recent clinical trials, current political events
Value-laden (ethics, aesthetics, contested social claims)	Produce a generic response or a "both sides" framing	Trained to avoid controversy; values questions have no settled answer in the training data	Ethics of genetic engineering

Discussion Guide

Activating findings (5 minutes): "Without looking at your notes — which AI answer do you trust more, and why? Biology or History?"

Building the framework (10 minutes): "Your intuition is correct — now let's explain it. It's not that biology is 'easier' or that AI was 'trying harder' for biology. What is DIFFERENT about the two subjects that explains the difference in AI reliability?" [Guide toward: the type of knowledge, whether experts agree, what the answer depends on]

Testing the framework (5 minutes): "Apply the framework to a subject we haven't discussed today. If you were asking AI 'Why is Shakespeare's Macbeth significant?', would you expect a reliable answer? What about 'What is the formula for calculating kinetic energy?' Use the framework to make a prediction — then we'll check."

The metacognitive close: "Before your next assignment, ask yourself: 'What type of knowledge am I asking AI about? Is this settled or contested?' That question takes 10 seconds and should determine how much you trust the answer."

---

Known Limitations

1. The vertical/horizontal distinction is not binary. Bernstein's categories are analytical — in practice, every discipline has both settled and contested knowledge. The framework should be taught as a useful heuristic, not a classification system. Biology has contested areas; history has settled facts. The framework predicts PATTERNS, not guarantees.

2. AI models vary in how well they handle contested claims. Some models are specifically trained to acknowledge debate and hedge their confidence in interpretive domains. The sequence may produce different findings with different AI tools. This is itself a useful finding — discussing why models handle these questions differently can deepen the AI literacy.

3. The sequence requires students to have sufficient domain knowledge to evaluate AI's answers. Students cannot assess whether AI's causal account of Germany's WWI defeat is oversimplified if they haven't studied WWI sufficiently. The sequence is most effective mid-unit or end-of-unit, after foundational knowledge is in place.

4. Direct empirical evidence for disciplinary AI literacy pedagogy is very limited. The knowledge-type framework (Bernstein, Willingham, Maton) is strongly evidenced for general curriculum design. Its specific application as a framework for predicting AI reliability is principled but novel — there is not yet substantial research on teaching students to predict AI reliability using knowledge-structure reasoning.

Prompt Literacy Sequence Designer

What This Skill Does

Generates a structured learning sequence that teaches students why prompt quality determines AI output quality — and what specific prompt moves produce more useful, accurate, and contextually appropriate responses. The sequence follows a compare-contrast structure: students run vague and refined prompts on the same question, analyse the difference in output quality, and abstract the principles. The core insight is that AI fills missing context with the most statistically common response — so a prompt with no context about audience, purpose, discipline, or constraints will receive an answer calibrated for the average case, not the student's specific situation. The Pricing Exercise (Kharbach, 2026) is included as the anchor activity: students take a context-free AI answer ("What should I charge for a service?") and iteratively add constraints (type of service, location, target market, quality level), showing in real time how specificity transforms output from generically unhelpful to genuinely useful. The sequence teaches five prompt dimensions: context (who am I, what am I doing?), task (exactly what do I want?), constraints (what limits apply?), format (how should the output be structured?), and persona (what role should the AI take?). Prompt literacy is a prerequisite for effective AI use and a direct complement to AI output evaluation skills.

Evidence Foundation

Brown et al. (2020) in the GPT-3 paper demonstrated empirically that the way a prompt is formulated dramatically affects model output quality — few-shot examples in the prompt (showing the AI what a good response looks like) produce substantially better results than zero-shot prompts (no examples). This is the foundational evidence that prompt design is not arbitrary. Liu et al. (2023) conducted a systematic survey of prompting methods, documenting the research on how different prompt structures (chain-of-thought, role-play, instruction-following, few-shot) affect output quality across tasks. Their survey establishes that prompt engineering is a skill with learnable principles, not a matter of chance. Reynolds & McDonell (2021) extended this to the concept of "metaprompts" — prompts that explicitly instruct the AI about how to reason, structure its response, or adopt a persona — showing that these structural elements can substantially improve output quality. These three sources provide the AI-specific evidence base for prompt literacy instruction. However, the pedagogical evidence base for teaching prompt literacy to students is currently very limited — this is frontier territory in educational research. The remaining sources support the instructional design of this sequence: Rosenshine (2012) provides the modelling → guided practice → independent practice structure used here; Willingham (2007) provides the domain-specificity argument (what counts as a good prompt in history is different from what counts as one in mathematics) that justifies subject-specific prompt literacy instruction.

Input Schema

The teacher must provide:

• Subject area: The discipline. e.g. "History" / "Biology" / "English Language & Literature" / "Mathematics"
• Student level: Year group and current AI usage. e.g. "Year 10, regularly use ChatGPT for homework but get generic outputs they don't find useful" / "Year 12, use AI for research and draft writing but haven't been explicitly taught prompt strategies"
• AI task type: What students use AI for. e.g. "Research: getting background information on essay topics" / "Writing: generating draft paragraphs and getting feedback" / "Explanation: asking AI to explain concepts they've missed"

Optional (injected by context engine if available):

• Prompt literacy focus: Which prompt dimension to emphasise
• Common student prompts: Real examples of how students currently prompt AI
• Time available: Duration for the sequence

Prompt

You are an expert in AI literacy pedagogy and instructional design, with knowledge of prompt engineering research (Brown et al. 2020; Liu et al. 2023; Reynolds & McDonell 2021) and instructional design principles (Rosenshine, 2012; Willingham, 2007). You understand that prompt literacy is a genuinely new skill with limited dedicated pedagogical research — the strongest evidence base is for the technical principles (prompt structure affects output quality) rather than for specific teaching methods. You will design a learning sequence grounded in instructional design principles applied to this new domain.

CRITICAL PRINCIPLES FOR PROMPT LITERACY:
- **AI fills missing context with the average.** When a student writes "explain photosynthesis," the AI generates an explanation calibrated for the most likely reader of that query — probably a general adult, not a Year 10 student preparing for a specific exam. Good prompts close the gap between the average case and the specific case.
- **Constraints are productive, not restrictive.** Most students think a prompt is "finished" when they've stated the task. Constraints — "in no more than 200 words," "without using the word 'significant'," "for an audience who has never studied biology" — transform output quality because they force the AI to solve a more specific problem.
- **Prompt literacy is discipline-specific.** What makes a good prompt for a History essay is different from what makes a good prompt for a Physics problem explanation. The principles are the same; the application differs.
- **Compare-contrast is the core learning mechanism.** Students learn prompt literacy most effectively by running two prompts side-by-side and analysing the difference — not by memorising rules. The rules become intelligible through contrast.
- **Prompt improvement is iterative, not one-shot.** Expert AI users refine their prompts based on the output they receive. Students need to understand that a first prompt is a starting point, not a final request.

Your task is to design a prompt literacy learning sequence for:

**Subject area:** {{subject_area}}
**Student level:** {{student_level}}
**AI task type:** {{ai_task_type}}

The following optional context may or may not be provided. Use whatever is available; ignore fields marked "not provided."

**Prompt literacy focus:** {{prompt_literacy_focus}} — if not provided, address all five prompt dimensions (context, task, constraints, format, persona) but weight the sequence toward the 2-3 most relevant for the stated AI task type.
**Common student prompts:** {{common_student_prompts}} — if not provided, generate realistic examples of how students at this level typically prompt AI for this task type — usually brief, task-only, no context.
**Time available:** {{time_available}} — if not provided, design for a 30-minute lesson or homework task.

Return your output in this exact format:

## Prompt Literacy Sequence: [Subject/Task Type]

**For:** [Student level]
**AI task type:** [What students are using AI for]
**Sequence focus:** [Which prompt dimensions are emphasised and why]

### Prompt Anatomy

[The five prompt dimensions with subject-specific examples for each:]

**1. Context:** [What context is — who I am, what I'm doing, why. Subject-specific example of no context vs. good context]
**2. Task:** [What task specificity means beyond just "explain X." Subject-specific example]
**3. Constraints:** [What constraints do — how they force more specific, useful output. Subject-specific example]
**4. Format:** [What format specification achieves. Subject-specific example]
**5. Persona:** [What role assignment achieves. Subject-specific example]

### Compare-Contrast Activity

**Vague prompt:** [A realistic example of how students currently write prompts for this task]

**What AI gives back:** [Description of what output this prompt generates — generic, unspecific, average-case]

**Refined prompt:** [The same task with all relevant dimensions added]

**What AI gives back:** [Description of how output quality changes — specific, audience-calibrated, more useful]

**Analysis questions for students:**
[4-5 questions that guide students to identify WHY the refined prompt produced better output — not just "it's better" but what specifically changed]

### The Pricing Exercise

[The core activity: a demonstration of how iterative constraint-adding transforms AI output.]

**The base prompt:** [A context-free question relevant to this subject area that produces a generically unhelpful answer]

**Step 1 — Add audience context:** [The prompt after adding who the user is and their purpose]
**Observed change:** [What changes in the output]

**Step 2 — Add constraints:** [The prompt after adding specific constraints relevant to the context]
**Observed change:** [What changes in the output]

**Step 3 — Add format specification:** [The prompt after adding output format requirements]
**Observed change:** [What changes in the output]

**The principle:** [One sentence: what this exercise demonstrates about the AI fill-with-average principle]

### Prompt Rewriting Activity

**Instructions:** [Step-by-step instructions for students to diagnose and rewrite a weak prompt]

**Weak prompt to rewrite:** [A realistic student prompt for this subject and task]

**Diagnosis framework:** [Questions students answer before rewriting: What context is missing? What constraints would improve this? What format would be most useful? What persona would help?]

**Student rewriting space:** [Guided template with slots for each prompt dimension]

**Share and compare:** [How students share their rewrites and compare outputs]

### Prompt Principles Summary Card

[A concise, printable reference for students — 5 principles with one-line explanations and short examples from this subject area]

**Self-check before returning output:** Verify that (a) examples are specific to the stated subject and AI task type, (b) the compare-contrast shows a concrete, observable difference in output quality, (c) the Pricing Exercise uses a realistic base prompt that genuinely produces a generically unhelpful answer, (d) the rewriting activity requires student judgment, not just slot-filling, and (e) the principles summary is genuinely memorable, not a list of abstract rules.

Example Output

Scenario: Subject area: "History" / Student level: "Year 11, use ChatGPT regularly for essay help but consistently receive generic outputs that don't address their specific exam requirements" / AI task type: "Writing: getting help with essay arguments and paragraph structure" / Common student prompts: "write me an essay about the causes of World War One" / "help me write a paragraph about the Treaty of Versailles"

---

Prompt Literacy Sequence: History Essay Writing (Year 11)

For: Year 11 History, experienced AI users getting generic outputs AI task type: Essay argument and paragraph help Sequence focus: Context (exam-specific framing) and constraints (argument structure, evidence requirements, word limits) — these two dimensions produce the greatest quality jump for this use case

Prompt Anatomy

1. Context: Without context, AI writes for the average reader of a WWI essay — a general adult. No context: "help me write a paragraph about the Treaty of Versailles." With context: "I'm a Year 11 student in the UK writing an AQA GCSE History essay. I need to argue that the Treaty of Versailles was the most important cause of World War Two."

2. Task: Stating the topic isn't enough — state the specific argument you need to build. Vague task: "explain the Treaty of Versailles." Specific task: "explain how the economic terms of the Treaty of Versailles contributed to political instability in Germany in the 1920s."

3. Constraints: Constraints turn an AI essay into a useable paragraph. Without constraints: you get 500+ words in any style. With constraints: "in approximately 120 words, structured as: claim, evidence, analysis, link to argument" — you get something usable.

4. Format: Specify structure. "Use the PEEL structure (Point, Evidence, Explanation, Link)" or "write it as bullet points I can expand myself, not a full paragraph" — these produce fundamentally different and more useful outputs.

5. Persona: "Act as a strict GCSE History examiner" produces different feedback than no persona. "Act as a student one year ahead of me who has got top marks" changes the register.

Compare-Contrast Activity

Vague prompt: "write me an essay paragraph about the Treaty of Versailles"

What AI gives back: A generic 200-word paragraph that explains what the Treaty of Versailles was, its terms, and that it "caused resentment in Germany." It has no specific argument, no GCSE exam structure, and would score poorly because it describes rather than analyses.

Refined prompt: "I'm writing a GCSE History essay arguing that the Treaty of Versailles was the most important cause of the rise of the Nazis. I need one well-structured paragraph (around 120 words) in PEEL format (Point, Evidence, Explanation, Link) that uses at least one specific piece of evidence from the treaty terms. Focus on the economic consequences rather than the territorial ones."

What AI gives back: A 120-word paragraph beginning with a clear historical argument, using the specific reparations figure (£6.6 billion) as evidence, explaining the mechanism (economic instability → unemployment → extremism), and linking explicitly to the essay thesis about Nazi rise. Structurally sound, exam-appropriate, immediately useful.

Analysis questions for students:

• "What information did I add in the refined prompt that the vague one was missing? List them."
• "Why did specifying 'economic consequences' rather than 'territorial ones' change what the AI wrote?"
• "The AI produced PEEL structure because I asked for it. What would have happened if I hadn't specified format?"
• "If the AI's first output was generic, was that the AI's fault or the prompt's fault? What does this tell you?"
• "How would you rewrite the refined prompt to get a different paragraph for the same essay?"

The Pricing Exercise

The base prompt: "What examples should I use to support my history essay argument?"

Observed change — after adding audience context: "I'm writing a Year 11 GCSE History essay on whether the Treaty of Versailles was the main cause of the rise of Nazism. What examples should I use to support this argument?" → AI now generates examples relevant to Weimar Germany and the Nazi rise, not random historical examples.

Step 2 — Add constraints: "...What 3 specific pieces of evidence should I use? Each should be a fact, statistic, or named event with a date." → AI produces three concise, specific, dateable pieces of evidence rather than a vague list of themes.

Step 3 — Add format specification: "...Present each piece of evidence as: [Evidence] — [Why this supports my argument]." → AI now produces a structured, immediately usable reference tool rather than flowing prose.

The principle: AI starts from the average case; every constraint you add moves the output closer to your specific situation.

Prompt Rewriting Activity

Instructions:

1. Read the weak prompt below.
2. Diagnose it using the four questions in the framework.
3. Rewrite it using the template — don't just make it longer, make each dimension more specific.
4. Compare your rewrite with a partner's.

Weak prompt to rewrite: "Explain the causes of World War One for my essay."

Diagnosis framework:

• What context is missing? (Who are you? What exam/task is this for? What argument are you building?)
• What constraints would improve this? (Word limit? Specific causes to focus on? Causes to exclude?)
• What format would be most useful? (List? PEEL paragraph? Bullet points for planning?)
• What persona would help? (Examiner giving feedback? Historian explaining the debate?)

Student rewriting template: "I am [context: who you are and what task this is for]. I need [specific task: exactly what you want, not just the topic]. Please [constraints: what to include/exclude, approximate length, specific angle]. Format your response as [format: structure you want]. [Optional: Approach this as a [persona]]."

Share and compare: In pairs, both students run their rewritten prompts and compare outputs. Discussion: Which rewrite produced a more useful response? Which of the five dimensions made the biggest difference? What would you add to your partner's rewrite?

Prompt Principles Summary Card

1. Give context first. Tell the AI who you are, what subject/level, and what the output is for. Without context, AI answers for nobody in particular.
2. Be specific about the task. "Explain X" is rarely enough. State the argument, the angle, or the specific question you need answered.
3. Add constraints. Word limits, required evidence, excluded approaches — constraints force specific, useful output.
4. Specify format. PEEL, bullet points, numbered list, dialogue — saying what structure you need takes 5 seconds and saves significant editing time.
5. Refine, don't restart. If the first output isn't quite right, respond with: "That's good but [specific issue]. Change [specific element]." Iterating is faster than rewriting from scratch.

---

Known Limitations

1. The evidence base for teaching prompt literacy to students is very limited. The research on prompt engineering (Brown et al. 2020; Liu et al. 2023; Reynolds & McDonell 2021) documents that prompt structure affects output quality — it does not study pedagogical approaches to teaching this skill. This sequence applies established instructional design principles (compare-contrast, modelling, guided practice) to a frontier domain. Teachers should treat it as principled but provisional.

2. Prompt literacy has a short shelf life as AI models improve. Current models require explicit context and constraints because they fill missing information with statistical averages. Future models may become better at inferring context, making some prompt strategies obsolete. The underlying principle (clear, specific communication produces better results than vague requests) is unlikely to become irrelevant; specific moves may change.

3. Teaching prompt literacy may increase AI dependence. Students who learn to write better prompts will get more useful AI outputs — which may reduce their motivation to develop independent skills. This skill should always be paired with ai-output-critical-audit-designer and metacognitive-monitoring-ai-contexts to ensure prompt literacy is part of a critical AI literacy framework, not a pure productivity optimisation.

4. Prompt outcomes are probabilistic, not deterministic. The same prompt can produce different outputs on different runs. Compare-contrast activities should acknowledge this — students may need to run prompts 2-3 times to see consistent patterns, and "the refined prompt is always better" is not quite accurate (it's usually better, in ways consistent with the principles, but not always).

What This Skill Does

This skill audits a curriculum framework against a statutory or accreditation requirement list and produces a structured coverage table showing which framework content addresses each requirement and how strongly. It is framework-agnostic: the framework being audited can be any school's curriculum, any programme of study, or any competency framework. The requirement list can be a statutory mandate, an accreditation standard, or any external checklist that the school is expected to meet. Neither input is assumed to be REAL School, UK, or any other specific context.

The skill classifies each requirement match as direct (the framework explicitly addresses the requirement), partial (the framework addresses related content but not the specific requirement), indirect (tangentially related content is present), or none (no framework content touches the requirement). This four-tier scale follows Webb's (1997) categorical-concurrence logic, operationalised at the requirement level rather than the item level. The skill is transparent about what "coverage" means in this context: topical presence in the framework documentation. A framework item that touches a requirement topic does not necessarily mean the requirement is fully taught, assessed, or enacted at the intended cognitive depth.

The skill produces three outputs: a Markdown coverage table for PLC or accreditation preparation, a matching CSV for spreadsheet filtering, and a prose gap summary for narrative reporting. The coverage_statistics object gives a headline count for quick scanning. These outputs are scaffolds for human professional judgement — they are designed to reduce the time a curriculum coordinator spends locating evidence, not to replace the judgement required to confirm it.

Evidence Foundation

Webb (1997) — The canonical four-dimension alignment framework establishes categorical concurrence as the foundational coverage dimension: do the framework and the requirement list address the same topics at all? The skill uses Webb's logic to classify match strength, stopping at categorical concurrence and range. Depth-of-knowledge alignment — whether the framework addresses a requirement at the required cognitive depth — is explicitly out of scope and flagged as a human judgement task in Known Limitations.

Porter et al. (2002) — The Surveys of Enacted Curriculum methodology establishes the row-per-requirement approach used in this skill's coverage table: each requirement gets its own row, and framework content is mapped to it rather than the reverse. This ensures that requirements with no framework coverage (the gap rows) are as visible as those with strong coverage, and prevents the natural tendency to organise the output around what the framework does cover rather than what the requirement list asks for.

Input Schema

Required inputs.

• framework — The curriculum framework being audited. Accepts band-tagged JSON from the Developmental Band Translator (for band-level filtering), or plain markdown or text with competencies and learning targets. If band-tagged, the skill can restrict the audit to specific bands via band_filter. If plain text, all framework content is audited without band distinction.
• requirements — The statutory or accreditation requirement list. Accepts markdown, CSV, or plain text. Each requirement should carry an identifier (number, code, or label) for the Requirement ID column. If identifiers are absent, the skill assigns stable sequential IDs (R001, R002, …) and notes this in the preamble.

Optional inputs.

• framework_name — Human-readable label used in output headings.
• requirements_source — Name of the statutory or accreditation body, used in output headings.
• band_filter — Restrict the audit to specific bands (e.g. "D,E"). Only relevant when framework is band-tagged JSON. If omitted, all framework content is searched.

Worked example of required inputs. REAL School Budapest auditing its wellbeing curriculum against the UK RSHE statutory requirements (July 2025): framework = the REAL School Wellbeing Framework (band-tagged or markdown); requirements = the UK DfE statutory RSHE requirement list; framework_name = "REAL School Budapest Wellbeing Framework"; requirements_source = "UK DfE Statutory RSHE, July 2025". The output would show each RSHE statutory bullet as a row, with REAL framework content (if any) mapped to it, classified by evidence strength.

Prompt

You are a curriculum coverage-audit assistant. Your task is to audit a curriculum framework against a requirement list and produce a structured coverage table, a gap summary, and coverage statistics.

Procedure

1. Read the requirements list. For each requirement, confirm it has a stable identifier. If identifiers are absent, assign sequential IDs (R001, R002, …) and note in your output preamble. Record every requirement before searching the framework — do not skip any.
2. Read the framework. If band-tagged, note the band labels and apply band_filter if supplied. If plain text, treat all content as a single pool.
3. For each requirement, search the framework for content that addresses it:
   • Direct — the framework explicitly and substantively addresses the requirement. The framework item's topic, framing, and intent match the requirement's topic, framing, and intent closely. A teacher could point to this item as evidence of coverage without qualification.
   • Partial — the framework addresses related content but not the specific requirement. The topic overlaps but the framing, depth, or scope differs. A teacher could point to this item as partial evidence but would need to add context.
   • Indirect — the framework contains content tangentially related to the requirement. The topic is adjacent but not the same. Evidence is inferential, not direct.
   • None — no framework content touches the requirement. This is a gap row.
4. Produce coverage_table — one row per requirement, with columns: Requirement ID | Requirement text | Covered by | Band(s) | Evidence strength | Notes.
5. Produce coverage_csv — the same data as coverage_table in flat CSV format. Column names exactly: requirement_id, requirement_text, covered_by, bands, evidence_strength, notes.
6. Write gap_summary — one prose paragraph summarising the requirements in the none and indirect rows. Group related gaps thematically if possible. State total count of none and indirect rows.
7. Produce coverage_statistics — counts: total_requirements, directly_covered, partially_covered, not_covered (none rows only). Include indirectly_covered as a separate count.

Classification discipline

When classifying evidence strength, err toward conservatism. A match that requires the auditor to make an inferential leap beyond what the framework text directly states should be classified as partial or indirect, not direct. The skill's job is to give an honest starting point for professional review, not to maximise the apparent coverage score. A framework that looks fully covered by aggressive classification is not useful to an accreditor or a curriculum coordinator trying to find real gaps.

Inputs

• framework: {{framework}}
• requirements: {{requirements}}
• framework_name: {{framework_name}}
• requirements_source: {{requirements_source}}
• band_filter: {{band_filter}}

Produce a JSON object with fields: coverage_table (markdown string), coverage_csv (CSV string), gap_summary (string), coverage_statistics (object).

Example Output

Context: REAL School Budapest Wellbeing Framework audited against UK RSHE statutory requirements, band_filter = "D,E".

coverage_table (8 rows shown):

Req ID	Requirement text	Covered by	Band(s)	Evidence strength	Notes
RSHE-D-01	"the characteristics of positive and healthy friendships… respect, honesty, trust, loyalty, kindness, generosity, boundaries, privacy, consent and the management of conflict"	"Students can identify healthy and unhealthy relationship patterns and articulate why each is so."	D	direct	Strong topical and intent match.
RSHE-D-02	"the concepts of, and laws relating to, sexual consent, sexual exploitation, abuse, grooming, coercion, harassment"	"Students understand consent as an ongoing, reversible, affirmative process."	D	partial	Framework addresses consent developmentally; statutory requirement includes legal framing and broader harm categories not explicitly present.
RSHE-D-03	"the facts about the full range of contraceptive choices, efficacy and options available"	—	—	none	No framework content addresses contraception at Bands D or E.
RSHE-D-04	"the impact of pornography on attitudes to sex, relationships and gender"	—	—	none	No direct equivalent in framework. Digital wellbeing content at Band E touches media influence but not pornography specifically.
RSHE-D-05	"how the different sexually transmitted infections (STIs), including HIV/AIDs, are transmitted, how risk can be reduced"	—	—	none	No STI content identified at Band D or E.
RSHE-E-01	"how to seek, and recognise, support for mental health and wellbeing"	"Students can identify when to seek help for themselves or a peer and know at least two trusted routes."	D	partial	Framework addresses this at Band D; statutory requirement is placed at End of Secondary. Band E has no equivalent item — partial coverage at wrong band.
RSHE-E-02	"the impact of viewing harmful content"	—	—	indirect	Digital wellbeing content at Band E covers technology use and wellbeing effects generally; harmful content is not specifically addressed.
RSHE-E-03	"I can analyse and evaluate how my identity, values and sense of self-worth affect the way I interact with others"	"Students can analyse how identity, values, and context shape their decisions about relationships."	E	direct	Close topical and intent match at the correct band.

coverage_csv (matching rows):

requirement_id,requirement_text,covered_by,bands,evidence_strength,notes
RSHE-D-01,"the characteristics of positive and healthy friendships… respect, honesty, trust, loyalty, kindness, generosity, boundaries, privacy, consent and the management of conflict","Students can identify healthy and unhealthy relationship patterns and articulate why each is so.",D,direct,Strong topical and intent match.
RSHE-D-02,"the concepts of, and laws relating to, sexual consent, sexual exploitation, abuse, grooming, coercion, harassment","Students understand consent as an ongoing, reversible, affirmative process.",D,partial,Framework addresses consent developmentally; statutory requirement includes legal framing not present.
RSHE-D-03,"the facts about the full range of contraceptive choices, efficacy and options available",,, none,No framework content addresses contraception at Bands D or E.
RSHE-D-04,"the impact of pornography on attitudes to sex, relationships and gender",,,none,No direct equivalent. Digital wellbeing content is adjacent but not specific.
RSHE-D-05,"how the different sexually transmitted infections (STIs), including HIV/AIDs, are transmitted",,, none,No STI content identified at Band D or E.
RSHE-E-01,"how to seek, and recognise, support for mental health and wellbeing","Students can identify when to seek help for themselves or a peer and know at least two trusted routes.",D,partial,Addressed at Band D not E; partial coverage at wrong band.
RSHE-E-02,"the impact of viewing harmful content",,,indirect,Digital wellbeing content is adjacent; harmful content not specifically addressed.
RSHE-E-03,"I can analyse and evaluate how my identity, values and sense of self-worth affect the way I interact with others","Students can analyse how identity, values, and context shape their decisions about relationships.",E,direct,Close topical and intent match.

gap_summary: Of 8 requirements sampled, 3 have no coverage (none) and 1 has only indirect coverage at Bands D and E. The none rows cluster around sexual health content: contraception options (RSHE-D-03), STI transmission and prevention (RSHE-D-05), and pornography and harmful content (RSHE-D-04 and RSHE-E-02, though the latter is indirect). These represent a substantive statutory gap at the secondary level. Additionally, the mental-health help-seeking requirement (RSHE-E-01) is partially met but at an earlier band than the statutory placement, meaning the Band E requirement is not addressed.

coverage_statistics:

{
  "total_requirements": 8,
  "directly_covered": 2,
  "partially_covered": 2,
  "indirectly_covered": 1,
  "not_covered": 3
}

Known Limitations

Topical coverage is not depth equivalence. A framework item that appears in the coverage_table as a "direct" match to a requirement is evidence that the topic is present in the framework documentation. It is not evidence that the topic is taught at the depth the requirement intends, assessed in the way the requirement expects, or enacted at the sophistication the accreditor would look for. Webb's (1997) depth-of-knowledge dimension — whether the cognitive demand of the framework content matches the cognitive demand of the requirement — is out of scope for this skill. A coverage_table row classified as direct is a starting point for professional review, not a conclusion.

Human professional judgement is required before any accreditation submission. This skill produces a scaffold for the curriculum coordinator's or accreditation reviewer's work. The coverage classifications are generated by pattern-matching against framework text; they have not been reviewed by a qualified professional in the relevant regulatory domain. Before using this output to support an accreditation claim, a competent professional must review each direct and partial row and confirm that the match holds under the accreditor's interpretation of the requirement. This skill explicitly does not substitute for that review.

The skill cannot assess taught, assessed, or enacted curriculum. The framework input describes the intended curriculum — what the school's documentation says will be taught. The skill audits documentation against documentation. It cannot determine whether the framework content is actually delivered in classrooms, whether it is assessed, or whether the enacted curriculum matches the written framework. Coverage in the table means coverage on paper. Accreditors and statutory inspectors typically require evidence of enacted curriculum, not just documented curriculum; this skill supports the documentation stage only.

What This Skill Does

This skill compares two or more band-tagged frameworks and produces a framework-neutral topic matrix as its primary output. Its input is the output of the Developmental Band Translator skill — one output per framework. Every framework is treated as an equal participant: topics present in any framework appear as rows in the matrix, and a framework's absence at a given theme × band is shown as an explicit "—" gap, not an omission. This makes the primary output useful for identifying what any framework lacks, not just where frameworks agree.

If a reference framework is also supplied, the skill additionally produces a secondary reference-centric crosswalk document — a Markdown document that a teacher or PLC facilitator can open and use directly for a professional discussion, organised under five required sections: Convergence, Divergence, Unique Content, Sequencing Differences, and Questions for PLC. The secondary document and its associated CSV are only produced when a reference framework is supplied.

The skill is deliberately additive, not reductive. It does not merge the frameworks into a single synthesised schema. It does not rewrite statements into a common voice. Each framework's content appears in the output in the original words the framework used, attributed by name. This is a design decision grounded in Martone & Sireci (2009): alignment work that hides framework distinctiveness loses the very information a PLC needs to make a planning decision.

The skill is opinionated about the distinction between "cross-framework agreement" and "cross-framework coincidence". Two frameworks addressing the same topic at the same band is convergence only if they also treat the topic at comparable depth and intent. Where that is unclear, the skill flags the pairing as apparent convergence, teacher confirmation needed rather than silently counting it as agreement.

Evidence Foundation

Webb (1997) — The four-dimension alignment framework (categorical concurrence, depth-of-knowledge consistency, range-of-knowledge correspondence, balance of representation) is the canonical reference point for alignment work in curriculum and assessment. This skill operationalises categorical concurrence (do the frameworks address the same topics?) and range (do the ages match?) at the school-band level. Depth-of-knowledge and balance are deliberately not assessed automatically; the Questions for PLC section surfaces them for human discussion.

Porter (2002) and Porter et al. (2007) — The Surveys of Enacted Curriculum methodology established that framework comparison requires a common content taxonomy. The theme taxonomy — whether supplied via theme_taxonomy or derived from all input frameworks — serves this role. The skill inherits the Porter methodology's commitment to making the comparison visible and defensible: every cell in every table traces back to specific items in specific frameworks.

Case, Jorgensen & Zucker (2004) — Practical alignment procedures in assessment explicitly recommend that alignment studies preserve framework differences rather than suppress them. The framework-neutral matrix exists for exactly this reason: a theme row where the reference framework cell is "—" is as informative as a row where it matches others. A crosswalk that showed only overlap would erase the most important planning signal.

Martone & Sireci (2009) — This review article establishes that alignment studies should preserve framework distinctiveness rather than collapse to a lowest-common-denominator schema. The skill's refusal to produce a merged mega-framework, and its commitment to showing each framework's content in its own voice, follow directly from this finding.

Input Schema

Required inputs. The skill needs one required input:

• comparison_frameworks_band_tagged — An array (two or more) of Developmental Band Translator outputs. All frameworks in this array are treated as equals in the primary matrix output. Beyond four frameworks the Convergence table in the secondary document becomes unwieldy and the skill emits a too_many_frameworks_for_readable_output flag.

Optional inputs.

• reference_framework_band_tagged — If supplied, the skill produces both the framework-neutral matrix AND the secondary reference-centric crosswalk_document and crosswalk_convergence_csv. The reference framework is included in the primary matrix as one column among equals. If omitted, only the framework-neutral matrix outputs are produced.
• reference_framework_name — A human-readable label for the reference framework used in output headings and table columns.
• theme_taxonomy — A pre-defined list of theme labels. If supplied, all framework content is mapped to these labels rather than themes being derived from the input. Use this when running the skill repeatedly across changing inputs to keep row labels consistent across runs.
• focus_bands — A comma-separated list of school bands to restrict the output to (e.g. "D,E" for a secondary PLC). Content outside these bands is excluded from all tables but acknowledged in preambles.
• focus_themes — Free-text theme restriction (e.g. "consent, relationships"). The skill filters items using keyword and semantic-similarity matching; a match log is retained in internal_trace.
• plc_context — One or two sentences describing the PLC audience and purpose. Used to tune the Questions for PLC section of the secondary crosswalk_document when a reference framework is supplied.

Worked example (reference framework supplied). A REAL School Budapest Band D/E wellbeing PLC preparing for September 2026 planning might pass: reference_framework_band_tagged = the REAL School Wellbeing Framework band-tagged output; comparison_frameworks_band_tagged = [band-tagged UK DfE Statutory RSHE, band-tagged Welsh CfW Health & Wellbeing]; reference_framework_name = "REAL School Budapest Wellbeing Framework"; focus_bands = "D,E"; plc_context = "Band D/E wellbeing team reviewing coverage gaps for 2026-27 planning". All three frameworks appear as equal columns in the primary matrix, with any REAL School gap cells shown as "—".

Worked example (no reference framework). A curriculum team comparing three published frameworks before designing a new scope and sequence might pass only comparison_frameworks_band_tagged = [band-tagged IB MYP Health, band-tagged UK RSHE, band-tagged Welsh CfW H&W]. The primary matrix shows all three on equal footing; no secondary crosswalk document is produced.

Prompt

You are a curriculum crosswalk assistant. You compare multiple band-tagged frameworks and produce a framework-neutral topic matrix as your primary output, and optionally a reference-centric crosswalk document as a secondary output.

Step 1 — Read all frameworks and extract all topics

For each framework in comparison_frameworks_band_tagged (and reference_framework_band_tagged if supplied), list every band-tagged item: school band(s), content statement, source-band label, and knowledge type if present. Produce a combined topic list spanning all frameworks. No topic is excluded at this step.

Step 2 — Group topics into coherent themes

If theme_taxonomy is supplied, map every topic from Step 1 to the nearest theme in that taxonomy. Do not invent new themes; flag unplaceable topics in theme_grouping_flags.

If theme_taxonomy is not supplied, derive themes from the full combined topic list. Group by semantic similarity and disciplinary coherence. Apply focus_themes filter (if supplied) after grouping.

For any grouping decision that is genuinely ambiguous — two frameworks addressing a topic at very different grain sizes, or content that plausibly belongs in two themes — add an entry to theme_grouping_flags. Each entry must include: the theme label, the source topics grouped under it, and the rationale for the grouping decision. Do not silently collapse ambiguous cases.

Step 3 — Build the theme × band × framework matrix

For each theme × band combination in scope (all bands, or focus_bands if supplied), find the relevant content statement from each framework. If a framework has no content for that theme × band, the cell is "—". An empty cell for the reference framework is a visible gap — do not omit it.

Record for each non-empty cell: the verbatim content statement, the source-band label (e.g. "End of Secondary", "Progression Step 4"), and a band_confidence value (high / medium / low) reflecting upstream ambiguity from the Developmental Band Translator.

Step 4 — Produce the matrix outputs

Produce three outputs:

framework_neutral_matrix — A Markdown table. Rows: theme × band. Columns: Theme, Band, then one column per framework (all on equal footing, including reference framework if supplied). Cells: verbatim content with source-band label in parentheses, or "—". Each "—" in any column is a visible gap.

framework_neutral_matrix_csv — A flat CSV at band level. One row per theme × band. Columns: theme, band, then for each framework: [framework_name]_content and [framework_name]_band_confidence, plus gap_count (integer: number of frameworks with "—" for this row), notes.

framework_neutral_summary_matrix — A summary CSV at theme level (one row per theme, not per band). Columns: theme, then for each framework: [framework_name]_covers (yes/no/partial) and [framework_name]_band_range (e.g. "D–E", "A–C", or "—").

Also produce theme_grouping_flags — the array of ambiguous grouping entries from Step 2.

Step 5 — Produce secondary outputs (only if reference framework supplied)

If reference_framework_band_tagged was supplied, produce the secondary reference-centric outputs using the following logic:

crosswalk_document — A Markdown document for direct use by a PLC facilitator. Contains a short preamble and these five sections:

1. Convergence table — content that the reference framework and at least one comparison framework address at the same school band. Columns: Band; Reference content (verbatim); Comparison framework; Comparison content (verbatim); Confidence (high / medium / apparent-only).
2. Divergence table — content that the reference framework and at least one comparison framework both address but at different school bands. Columns: Topic; Reference band; Reference content (verbatim); Comparison framework; Comparison band; Comparison content (verbatim); Source label; Band-gap (in band units).
3. Unique-content table — content present in only one framework. One subsection per framework.
4. Sequencing-difference notes — prose description of meaningful differences in how the frameworks sequence related content. Focus on differences with planning implications.
5. Questions for PLC — 6–10 numbered questions grounded in specific cells from the tables above. Do NOT answer them.

crosswalk_convergence_csv — A flat CSV. One row per LT × band × comparison framework pairing. Columns: lt_id, lt_name, band, reference_content, comparison_framework, comparison_content, comparison_source_label, confidence, issue_type, notes. Every convergence, divergence, and unique-content row appears. Empty comparison_content is data, not an omission.

If reference_framework_band_tagged was not supplied, omit both crosswalk_document and crosswalk_convergence_csv entirely.

Step 6 — Produce skill_flags and internal_trace

skill_flags — Emit warnings for: insufficient overlap to produce meaningful convergence; framework with >50% upstream ambiguity; focus_bands excluding all content in one framework; more than four frameworks producing unwieldy secondary output; fewer than 3 items remaining after filtering in any framework.

internal_trace — Framework ids compared, item counts per section, band-coverage matrix, theme derivation log (or taxonomy mapping log if theme_taxonomy was supplied), focus filter match log.

Output format (hard constraints)

• framework_neutral_matrix and crosswalk_document: Markdown only. No JSON, no YAML, no field names, no {variable} placeholders.
• All framework content quoted verbatim. Do not paraphrase, shorten, or translate content statements. Quote marks around exact source text.
• Attribute every content statement to its source framework by name.
• Preserve source-band labels alongside school bands in every table row.
• If an item's band assignment was ambiguous upstream (multi-band), list every relevant band, not just one.

Source-voice preservation (hard constraint)

Every content statement you reproduce must appear exactly as it did in the band-tagged input. Quote it. Attribute it. Do not rewrite it to match another framework's style. This is not stylistic — it is the crosswalk's core integrity property.

Inputs

• reference_framework_band_tagged: {{reference_framework_band_tagged}}
• reference_framework_name: {{reference_framework_name}}
• comparison_frameworks_band_tagged: {{comparison_frameworks_band_tagged}}
• theme_taxonomy: {{theme_taxonomy}}
• focus_bands: {{focus_bands}}
• focus_themes: {{focus_themes}}
• plc_context: {{plc_context}}

Produce framework_neutral_matrix, framework_neutral_matrix_csv, framework_neutral_summary_matrix, and theme_grouping_flags in all cases. Produce crosswalk_document and crosswalk_convergence_csv only if reference_framework_band_tagged was supplied. Always produce skill_flags and internal_trace.

Example Output

Context: Band D/E wellbeing PLC, comparing REAL School Budapest Wellbeing Framework (reference, supplied via reference_framework_band_tagged) against UK DfE Statutory RSHE and Welsh CfW Health & Wellbeing (in comparison_frameworks_band_tagged). focus_bands = "D,E".

---

(a) framework_neutral_matrix_csv — 5-row sample

Note: the REAL School cell is "—" in row 4 (STI/Contraception), making the gap explicit.

theme,band,REAL_School_Wellbeing_content,REAL_School_Wellbeing_band_confidence,UK_RSHE_content,UK_RSHE_band_confidence,Welsh_CfW_content,Welsh_CfW_band_confidence,gap_count,notes
Consent Understanding,D,"Students understand consent as an ongoing, reversible, affirmative process.",high,"the concepts of, and laws relating to, sexual consent, sexual exploitation, abuse, grooming, coercion",medium (End of Secondary spans C–E),—,—,1,Welsh CfW has no direct consent content at PS3/PS4
Mental-health Help-seeking,D,"Students can identify when to seek help for themselves or a peer and know at least two trusted routes.",high,"where to get advice e.g. family, school and/or other sources",low (End of Primary — band D is out of RSHE's intended range),—,—,1,RSHE places this 2 bands earlier; placed here for nearest-band comparison
Digital Wellbeing,E,"Students can evaluate the wellbeing effects of their own technology use and adjust it.",high,—,—,"I can make informed decisions about how to use digital technologies… to enhance my physical health and well-being",medium (PS3 spans C–D),1,Welsh CfW places this 1–2 bands earlier; no RSHE coverage at E
STI and Contraception,E,—,—,"the facts about the full range of contraceptive choices, efficacy and options available",high (End of Secondary),—,—,2,Reference framework gap — REAL School has no content matching this theme at Band E
Natural Environment and Wellbeing,D,—,—,—,—,"I can explain the benefits of spending time in the natural environment for my health and well-being",high (PS3),2,Reference framework gap — theme absent in both REAL School and UK RSHE at Band D

---

(b) framework_neutral_summary_matrix — 3-row sample

theme,REAL_School_Wellbeing_covers,REAL_School_Wellbeing_band_range,UK_RSHE_covers,UK_RSHE_band_range,Welsh_CfW_covers,Welsh_CfW_band_range
Consent Understanding,yes,D,yes,C–E (End of Secondary),no,—
STI and Contraception,no,—,yes,C–E (End of Secondary),no,—
Natural Environment and Wellbeing,no,—,no,—,yes,C–D (PS3)

---

(c) theme_grouping_flags — 2-entry sample

[
  {
    "theme": "Consent Understanding",
    "source_topics_grouped": [
      "REAL School: 'Students understand consent as an ongoing, reversible, affirmative process.'",
      "UK RSHE: 'the concepts of, and laws relating to, sexual consent, sexual exploitation, abuse, grooming, coercion, harassment, rape'"
    ],
    "rationale": "Both address consent but at different epistemic grains and framing — REAL School frames consent developmentally; UK RSHE frames it propositionally and in law. Grouped under one theme for matrix comparability, but these may represent genuinely distinct learning goals. Flagged for human review."
  },
  {
    "theme": "Digital Wellbeing",
    "source_topics_grouped": [
      "REAL School: 'Students can evaluate the wellbeing effects of their own technology use and adjust it.'",
      "Welsh CfW: 'I can make informed decisions about how to use digital technologies… to enhance my physical health and well-being'"
    ],
    "rationale": "Both concern technology and wellbeing, but REAL School frames this as a self-regulation competency while Welsh CfW frames it as a health-literacy decision skill. Grouped under one theme but the pedagogical intent differs. A school treating these as one topic may be collapsing a meaningful distinction."
  }
]

---

(d) Secondary output note

crosswalk_document is the secondary PLC-ready Markdown document, produced only when reference_framework_band_tagged is supplied. When produced, it follows the same five-section structure as in version 1.0 (Convergence, Divergence, Unique Content, Sequencing Differences, Questions for PLC), using REAL School as the reference column. Example content for that document follows the same format as the version 1.0 example — see that section's sample tables for illustration. The key difference is that the document is now secondary: the framework-neutral matrix is always produced first and is the primary analytical artefact.

Known Limitations

The skill does not assess depth-of-knowledge alignment. Webb's depth-of-knowledge dimension is out of scope for automated comparison. The skill can see that two frameworks both address "consent" at the same band, but it cannot reliably judge whether they do so at the same cognitive depth. The Questions for PLC section is where depth-of-knowledge discussions belong. Convergence at apparent-only confidence is explicitly flagged for this reason.

Theme derivation is the most judgement-dependent step. Grouping topics from multiple frameworks into coherent themes is interpretive, not mechanical. Two frameworks addressing "self-regulation" at very different grains may or may not belong in the same theme row — the skill makes an interpretive proposal, not an authoritative classification. theme_grouping_flags surfaces uncertain cases for human review rather than hiding them. Use the theme_taxonomy input for consistent row labels across repeated runs on changing inputs.

The skill inherits upstream ambiguity. Where the Developmental Band Translator tagged items with multiple candidate school bands (common for UK RSHE's 2-band structure), the matrix carries that ambiguity into every cell. A framework with high upstream ambiguity produces a matrix with many multi-band rows, which is honest but can be harder to read. The skill emits a skill flag when >50% of upstream items are multi-band.

The skill does not merge frameworks. This is intentional (Martone & Sireci, 2009). The skill will not produce a synthesised unified framework, nor a single recommended topic order. Those are planning decisions informed by this output but made by humans.

Thematic matching is not perfect. The skill pairs items by semantic similarity in content statements. False negatives (genuine matches missed because the language is very different) and false positives (superficial lexical overlap with different pedagogical intent) are both possible. The apparent-only confidence tier surfaces false-positive risk; theme_grouping_flags and the Questions for PLC section backstop false-negative risk. The skill is a scaffold for the teacher's reading of the source documents, not a substitute for it.

What This Skill Does

This skill tags an external curriculum — already decomposed by the curriculum harness into KUD items, learning targets, and (optionally) criteria — with a school's developmental band labels. The school's band schema is supplied as an input; the skill reads it, derives a mapping table from the source's age ranges to the school's band ranges, and applies that mapping across every item. Every item receives a school_band value, a confidence level, and diagnostic flags. The source's own band labels are preserved verbatim alongside the new school band tags, so the output can always be traced back to, and interpreted against, the original framework.

The skill is a metadata operation, not a content operation. It does not rewrite KUDs, LTs, or criteria, and it does not reinterpret the source's intent. Its only job is to answer the question: "If a teacher planning for a given school band opened this framework, which of its content is relevant to them?" The skill is deliberately conservative where a source band spans two school bands: it assigns both candidate bands, sets an ambiguity flag, and records a rationale. Downstream skills — especially the Curriculum Crosswalk skill and the human planning process — resolve these ambiguities with teacher judgement.

The skill handles three common source-structure scenarios. Most sources have an explicit band structure (Welsh CfW Progression Steps, NZ Curriculum Levels, IB PYP/MYP/DP phases) which maps cleanly with occasional boundary ambiguity. Some have a formal-but-compressed structure (UK RSHE has four Key Stage labels but only two actual age bands in the statutory document), where the skill applies the source's real 2-band structure first and records the formal label via source_band_preserved. A minority have no band structure at all (CASEL, some competency frameworks), where the skill maps by developmental descriptor rather than age and flags every item for teacher review.

Evidence Foundation

Wiggins & McTighe (2005, 2011) — Understanding by Design establishes KUD (Know / Understand / Do) as the unit-level architecture through which framework content becomes teachable, and positions curriculum alignment as a backward-design operation from stated outcomes. The Developmental Band Translator operates on exactly the KUD layer that UbD specifies, and treats school-band assignment as one step in a backward-design pipeline whose final consumer is the classroom teacher.

Vygotsky (1978) — The Zone of Proximal Development grounds the concept that developmental banding is not a pure age-partition. A learner in a given band is defined by what they can do with support, not only by their chronological age. This justifies the skill's use of both age-range and developmental-descriptor evidence when tagging, and its willingness to assign a school band from descriptor-based reasoning alone when age evidence is absent.

Heritage (2008) — Learning progressions literature reinforces that progressions are developmental trajectories, not calendars. The skill treats school bands as trajectory-waypoints. Where a source's native progression is coarser than the school's (e.g. RSHE's 2-band structure vs a 6-band school schema), the skill preserves the coarseness honestly via ambiguity flags rather than forcing false precision.

Webb (1997) — The canonical alignment framework defines four alignment dimensions: categorical concurrence, depth-of-knowledge consistency, range-of-knowledge correspondence, balance of representation. Band translation addresses primarily the range dimension (do the age spans match?) with partial touch on categorical concurrence. The skill does not claim to assess depth or balance alignment; this limit is reported in skill_flags and in Known Limitations.

Input Schema

Required inputs. The skill needs four inputs. The age-context rules below clarify the legacy age-range descriptions in the unchanged frontmatter: check a supplied override and the flat-framework exception before requesting missing native age ranges.

• band_schema — The school or programme's developmental band definitions. Must include band labels, approximate age or grade ranges, and any named groupings. The skill reads this to build the mapping table; it does not assume any particular band system. Example: "REAL School Budapest uses Bands A–F — A (Water+Air Dragons, K–2, ages 5–7), B (Earth Dragons, G3–4, ages 7–9), C (Fire Dragons, G5–6, ages 9–11), D (Metal+Light Dragons, G7–8, ages 11–13), E (G9–10, ages 13–15), F (G11–12, ages 15–17)."
• kud_json — The KUD file. Required fields per item: item_id, content_statement, knowledge_type, source_block_id. The skill uses source_block_id to look up the block's native band label in the progression structure.
• lts_json — The LT file. Required fields per LT: lt_id, lt_name, lt_definition, kud_item_ids, knowledge_type. LTs inherit their band from their constituent KUD items using a max-span rule (see Prompt).
• progression_structure_json — The progression structure. Required: band_labels and source_type; use a top-level age_range_hint or per-band approximate_age_range for age-based mapping. If native ages are absent, first check user_age_range_override and the flat-framework branch below. Request age context and halt only when neither provides a defensible mapping.

Optional inputs.

• criterion_bank_json — If supplied, criteria are also tagged. Criteria inherit their parent LT's band unless their content narrows to a more specific age-range signal.
• source_name — A friendly name for output metadata.
• user_age_range_override — A user-supplied age-range mapping; it may be supplied on the first invocation or after request_age_range_context. Apply it as mapping context before the no-age check, preserving native labels and recording its user-supplied status and any conflict with native ages in skill_flags.

Worked example of required inputs. For a REAL School Budapest run against uk-statutory-rshe: band_schema = "REAL School Budapest uses Bands A–F — A (Water+Air Dragons, K–2, ages 5–7), B (Earth Dragons, G3–4, ages 7–9), C (Fire Dragons, G5–6, ages 9–11), D (Metal+Light Dragons, G7–8, ages 11–13), E (G9–10, ages 13–15), F (G11–12, ages 15–17)"; kud_json = the 279-item KUD file; lts_json = the 44-LT file; progression_structure_json = a structure declaring band_labels: ["End of Primary (Year 6)", "End of Secondary (Year 11)"], source_type: "national_statutory_curriculum", and age ranges 5–11 and 11–16 respectively.

Prompt

You are a curriculum band-translation assistant. Your task is to tag an external framework — already decomposed by the curriculum harness — with a school's developmental band labels, while preserving the source's own voice and labels completely unchanged.

School band schema

Read the supplied band_schema carefully. It defines the school's band labels, grade spans, age ranges, and any named groupings. All band assignments you produce must use the band labels exactly as defined in that schema.

Procedure

1. Read the progression structure and choose mapping context. Preserve source_type and band_labels. First apply an explicit user_age_range_override as user-supplied mapping context, without rewriting native labels or ages; flag any conflict with native ranges for teacher review. Otherwise use native age_range_hint or per-band approximate_age_range. If neither exists and this is a competency/dispositional framework with no native age structure, use the conservative flat-framework branch below. Only if none of these routes supports mapping, emit request_age_range_context in skill_flags, do not tag anything, and stop. Do not invent ages.
2. Build a band-mapping table for this source. Using the band_schema and the mapping context selected in step 1 (or the flat-framework descriptors), identify which school bands overlap with each source band. Where a source band's age range spans multiple school bands, list all overlapping school bands as candidates and note the ambiguity. Where the source band maps cleanly to one school band, record low ambiguity. Record this mapping table in your reasoning before proceeding; every tagging decision in steps 3–5 must trace back to it.
3. Tag every KUD item.
   • Look up its source_block_id to find the source band.
   • Apply the band-mapping table to get candidate school bands.
   • school_band: single band as a string if unambiguous; array of bands if ambiguous.
   • band_confidence: high if single school band and source age range ≤ the school's narrowest band span; medium if two school bands; low if three or more, or if derived by developmental-descriptor inference only.
   • source_band_preserved: the exact native label from the source (e.g. "Key Stage 2", "End of Primary", "Progression Step 3").
   • source_voice_preserved: always true. If you would need to rewrite the content to make it fit a school band, do NOT rewrite — instead set this to false and add a skill flag. This field exists to prove no rewriting occurred.
   • ambiguity_flag: true if school_band is an array.
   • teacher_review_flag: true when (a) school_band spans three or more bands, (b) confidence is low, (c) source_type is a dispositional/competency framework with no age-banding (e.g. CASEL), or (d) the content statement contains developmental language that would narrow the band beyond what the source label allows, or (e) a user-supplied override conflicts with native age ranges.
   • band_rationale: one sentence explaining the assignment.
4. Tag every LT. An LT's school band is the union of its constituent KUD items' bands (max-span rule). If any constituent is ambiguous, the LT is ambiguous. teacher_review_flag fires if any constituent has it or if the LT aggregates across three or more school bands.
5. Tag criteria (if criterion_bank_json supplied). Each criterion inherits its parent LT's band unless its content narrows it. Same field set as LTs.
6. Emit summary counts and skill flags. Count items per school band, ambiguity flags, teacher-review flags, and low-confidence tags. Emit a skill flag for any source where >50% of items are ambiguous (signals a structural mismatch worth reviewing).

Source-voice preservation rule (hard constraint)

You are tagging content, not rewriting it. Do not paraphrase, shorten, or "translate" content statements, LT definitions, or criterion descriptions into the school's preferred prose style. Keep source content exactly as written. The only new text you generate is band_rationale.

CASEL / flat-framework handling

If source_type indicates a competency or dispositional framework with no age structure (e.g. CASEL), and no user_age_range_override was supplied:

• Set teacher_review_flag: true on every item. If a descriptor does not support narrowing, retain all plausible school bands, explain the uncertainty, and ask for teacher context rather than guessing a precise age.
• Set band_confidence: low on every item.
• Base school_band on the developmental sophistication of the descriptor, mapped to the school bands from the supplied band_schema: early-developmental content (emotion-labeling, basic self-regulation) → earliest bands in the schema; mid-developmental content (perspective-taking, collaborative problem-solving) → mid-range bands; late-developmental content (systems-level social awareness, abstract ethical reasoning) → later bands in the schema.
• Emit skill_flag: casel_style_framework_full_teacher_review_required.

Inputs

• band_schema: {{band_schema}}
• source_name: {{source_name}}
• progression_structure_json: {{progression_structure_json}}
• kud_json: {{kud_json}}
• lts_json: {{lts_json}}
• criterion_bank_json: {{criterion_bank_json}}
• user_age_range_override: {{user_age_range_override}}

Produce the band-tagged output as a single JSON object conforming to the output schema. Do not rewrite source content. Do not invent bands the source cannot support.

Example Output

Context: REAL School Budapest band schema supplied as band_schema. Source: UK DfE Statutory RSHE (July 2025), progression_structure.band_labels = ["End of Primary (Year 6)", "End of Secondary (Year 11)"], source_type = "national_statutory_curriculum".

Illustrative input excerpt (the required KUD and LT files are omitted here for space; they must be supplied in a real run):

{
  "band_schema": "REAL School Budapest uses Bands A–F — A (Water+Air Dragons, K–2, ages 5–7), B (Earth Dragons, G3–4, ages 7–9), C (Fire Dragons, G5–6, ages 9–11), D (Metal+Light Dragons, G7–8, ages 11–13), E (G9–10, ages 13–15), F (G11–12, ages 15–17). Bands are defined primarily by grade level; ages are approximate.",
  "source_name": "UK DfE Statutory RSHE, July 2025",
  "progression_structure_json": "{ \"band_labels\": [\"End of Primary (Year 6)\", \"End of Secondary (Year 11)\"], \"source_type\": \"national_statutory_curriculum\", \"age_range_hint\": { \"End of Primary (Year 6)\": \"5–11\", \"End of Secondary (Year 11)\": \"11–16\" } }"
}

Output excerpt (summary counts below cover only these three KUD items and one LT):

{
  "source_metadata": {
    "source_name": "UK DfE Statutory RSHE, July 2025",
    "source_type": "national_statutory_curriculum",
    "source_band_labels": ["End of Primary (Year 6)", "End of Secondary (Year 11)"],
    "skill_version": "1.0",
    "run_timestamp": "2026-04-20T14:05:00Z"
  },
  "band_tagged_kud": [
    {
      "item_id": "blk_0002_item_01",
      "content_statement": "Young people understand the correct terms for different parts of the body",
      "knowledge_type": "Type 1",
      "school_band": ["A", "B", "C"],
      "band_confidence": "low",
      "source_band_preserved": "End of Primary (Year 6)",
      "source_voice_preserved": true,
      "ambiguity_flag": true,
      "teacher_review_flag": true,
      "band_rationale": "Source groups all primary-age content into one band spanning ages 5–11; REAL Bands A–C cover this same range (5–7, 7–9, 9–11 respectively)."
    },
    {
      "item_id": "blk_0346_item_03",
      "content_statement": "Pupils can communicate their boundaries about physical contact and privacy needs",
      "knowledge_type": "Type 2",
      "school_band": ["A", "B", "C"],
      "band_confidence": "low",
      "source_band_preserved": "End of Primary (Year 6)",
      "source_voice_preserved": true,
      "ambiguity_flag": true,
      "teacher_review_flag": true,
      "band_rationale": "Statutory End-of-Primary expectation covering ages 5–11; maps to REAL Bands A–C."
    },
    {
      "item_id": "blk_0210_item_02",
      "content_statement": "Students understand the concepts of, and laws relating to, sexual consent",
      "knowledge_type": "Type 1",
      "school_band": ["D", "E", "F"],
      "band_confidence": "low",
      "source_band_preserved": "End of Secondary (Year 11)",
      "source_voice_preserved": true,
      "ambiguity_flag": true,
      "teacher_review_flag": true,
      "band_rationale": "End-of-Secondary covers ages 11–16; REAL Bands D (11–13), E (13–15), and F (15–17) overlap this range; the partial overlap with F is retained for teacher review."
    }
  ],
  "band_tagged_lts": [
    {
      "lt_id": "cluster_01_lt_01",
      "lt_name": "Identifying Body Parts and Privacy",
      "school_band": ["A", "B", "C"],
      "band_confidence": "low",
      "source_band_preserved": "End of Primary (Year 6)",
      "source_voice_preserved": true,
      "ambiguity_flag": true,
      "teacher_review_flag": true,
      "band_rationale": "Aggregated from 4 KUD items all tagged End of Primary; max-span = REAL Bands A–C."
    }
  ],
  "summary_counts": {
    "total_kud_items": 3,
    "total_lts": 1,
    "kud_by_band": {"A-B-C_ambiguous": 2, "D-E-F_ambiguous": 1},
    "lts_by_band": {"A-B-C_ambiguous": 1},
    "teacher_review_flagged": 4,
    "low_confidence": 4
  },
  "skill_flags": [
    "high_ambiguity_rate_structural: 100% of items are ambiguous because source has 2 bands spanning 3 school bands each. This is expected for national_statutory_curriculum sources with compressed band structures; do not treat as a quality failure. Downstream planning must resolve via teacher judgement."
  ]
}

Known Limitations

The skill does not do construct alignment. Two items tagged to the same school band are not, by that fact alone, teaching the same thing. Webb's (1997) categorical-concurrence and depth-of-knowledge dimensions are out of scope. The skill answers "is this age-relevant?" not "does this match the school's learning intentions?" The Curriculum Crosswalk skill addresses relationships between frameworks but still requires human judgement for construct alignment.

Ambiguity is not resolvable by this skill. When a source's band structure is coarser than the school's — particularly UK RSHE's 2-band structure or any framework with fewer bands than the school schema — the skill deliberately preserves the ambiguity rather than guessing. High ambiguity rates (up to 100% for RSHE against a 6-band schema) are an honest report of the source's granularity, not a quality failure. They are resolved downstream by teacher judgement, which is why teacher_review_flag exists.

Developmental-descriptor inference (CASEL case) is approximate. When a framework has no age-banding, the skill maps by descriptor using the supplied band_schema as a guide — early-developmental descriptors → earliest bands in the schema; systems-level descriptors → later bands. These are provisional descriptor-based starting points, not validated age norms or framework-specific determinations. Content tagged this way should be treated as a scaffold for teacher review, not a validated mapping.

The skill assumes harness-validated input. It does not re-validate the source decomposition. If the upstream harness produced a broken KUD or LT file (cycles in prerequisite graph, content_statement missing, etc.), the skill will pass those errors through as band-tagged nonsense. Run the harness's own validation gates before invoking this skill.

What This Skill Does

This skill authors or reviews Know/Understand/Do (KUD) charts for learning targets in competency-based developmental band frameworks. A KUD chart is a planning and teaching tool — not an assessment structure — that specifies what students need to factually hold (Know), what transferable conceptual insight they should develop (Understand), and what they will produce, perform, or demonstrate as evidence of capability (Do). The skill handles seven distinct starting points, from a raw government curriculum document to an existing KUD chart needing quality improvement, and routes to upstream skills when better-prepared inputs would produce stronger output.

The skill enforces two critical distinctions that are commonly collapsed in practice. First, the Know and Understand layers are teaching context, not additional assessment instruments. A student is assessed against the Do layer; the K and U layers inform how learning is designed and sequenced. This means a single learning target can have an analytically rich Understand layer without becoming a compound LT requiring a split. Second, Do cells are typed as either performance (student produces or executes a discrete evaluable artefact — essay, presentation, seminar contribution, lab report) or disposition (student demonstrates a behavioural pattern across time and contexts that cannot be adequately evidenced in a single occasion). Most T1 and T2 LTs have performance Dos assessed by rubric. Most T3 LTs have dispositional Dos assessed through multi-informant observation. The skill makes this distinction explicit in every Do cell.

The skill also types prerequisite relationships — distinguishing hard prerequisites (logically impossible to proceed without; warrants a readiness check) from soft enablers (enriches and accelerates but does not gate assessment) from conceptual accelerators (makes a competency portable and transferable to novel contexts). Most knowledge-to-disposition relationships are soft enablers, not hard gates. Over-gating — treating enrichment relationships as hard prerequisites — causes teachers to delay dispositional work unnecessarily.

Evidence Foundation

Wiggins & McTighe (2005, 2011) — Understanding by Design establishes KUD as the unit-level planning architecture through which curriculum intent becomes teachable. The critical UbD insight operationalised here is that Know, Understand, and Do are qualitatively different things requiring different instructional and assessment approaches. An Understand is not a fact to reproduce; it is a transferable idea a student carries into new situations. A Do is not a topic to cover; it is an observable demonstration of capability.

Black & Wiliam (1998) — Inside the Black Box establishes the theoretical basis for treating the K and U layers as teaching infrastructure rather than assessment targets. Formative assessment operates on the gap between current and desired performance (Do), informed by what students need to hold (Know) and grasp (Understand). Conflating the teaching layer with the assessment layer produces rubrics that assess the wrong thing.

Anderson & Krathwohl (2001) — The revised Bloom's taxonomy provides the cognitive science basis for the Know/Understand distinction. Know maps to remembering (retrieving relevant knowledge from long-term memory); Understand maps to understanding (constructing meaning from instructional messages). These are genuinely different cognitive operations requiring different instructional approaches and different assessment instruments.

Bandura (1997) — Self-Efficacy: The Exercise of Control provides the theoretical basis for the performance vs disposition distinction in Do cells. Dispositional capabilities — self-regulation, empathy, metacognitive self-direction — develop through repeated enactment across contexts over extended periods. They are not demonstrated in a single performance occasion. Assessment of dispositions requires accumulated evidence from multiple informants across time, not a rubric applied to a single task.

Heritage (2008) — Learning Progressions reinforces that developmental progressions must describe what learners can do with content across bands, not just harder content. Progression levers (independence, complexity, scope, precision, reasoning, transfer) are the mechanism by which this skill ensures band-to-band transitions are genuinely developmental rather than merely topical.

Input Schema

The skill handles seven input types. Supply the most complete input available for your situation — the routing note will flag if a better upstream starting point exists.

existing_lt_with_do — You have Do statements or observation indicators already written for each band but no K or U layers. The skill authors K and U to complement the existing Do without rewriting it. Most common use case when upgrading a working draft framework. Example: supply the existing band statement table from your LT spreadsheet.

competency_only — You have a competency definition and programme purpose but no LTs yet. The skill authors a full KUD from scratch. Note: running the Learning Target Authoring Guide skill first to establish the Do layer will produce a stronger result — the routing note will flag this.

harness_output — You have structured KUD/LT output from the curriculum harness pipeline (JSON or markdown). The skill translates harness output into a programme-formatted KUD chart, applying the performance vs disposition distinction and prerequisite typing.

crosswalk_data — You have relevant rows from a framework-neutral crosswalk matrix showing what comparable frameworks do at each band for this theme. The skill uses this as developmental scaffolding for the K and U layers without copying external framework voice.

unit_plan — You have a unit plan document. The skill extracts implicit K/U/D content from the unit plan and organises it into a KUD chart, then flags gaps.

existing_kud_upgrade — You have an existing KUD chart assessed as below quality bar. The skill runs all six quality checks and proposes specific cell-level improvements. Use mode: review.

raw_curriculum_document — You have a government curriculum document, exam specification, or course outline not yet decomposed. The skill reads the source and authors a KUD directly. Note: for documents over approximately 10 pages, first decomposing it through the curriculum harness can provide stronger structured inputs. Only then can the Developmental Band Translator add band metadata; it does not decompose raw documents. The routing note will flag this, and the skill will proceed with a draft regardless.

Prompt

You are a curriculum design specialist authoring Know/Understand/Do (KUD) charts for competency-based learning targets in a developmental band framework. You apply the Understanding by Design methodology (Wiggins & McTighe, 2005) with strict discipline about what KUD is for: a planning and teaching tool, not an assessment structure.

---

STEP 0 — ROUTING CHECK

Read {{input_type}} and {{primary_input}}. Before authoring, produce a routing_note:

If input_type is raw_curriculum_document AND the document appears longer than approximately 10 pages or structurally complex: set routing_note to — "This document would benefit from curriculum-harness decomposition into KUD items, LTs and a progression structure before band tagging. The Developmental Band Translator requires band_schema, kud_json, lts_json and progression_structure_json; it cannot accept this raw text as their substitute. Proceeding with a direct draft — expect the Know layer and progression levers to need more teacher review than usual."

If input_type is competency_only AND no existing_do_statements are supplied: set routing_note to — "Running the Learning Target Authoring Guide skill first will establish the Do layer. KUD authoring is stronger when the Do layer exists to work around. Proceeding with a full-authoring draft."

In all other cases: set routing_note to null.

Always proceed with authoring regardless of the routing note. Never block on the routing check. For raw documents, first identify source sections and distinguish explicit K/U/D content from inferred draft content in authoring_notes; retain traceable section references without importing school-specific programme names into the chart. This direct draft is not a claim that harness JSON has been produced or validated. Recommend a band-tagger handoff only after the required structured files and school band schema actually exist. The harness is an upstream workflow, not a newly assumed library skill.

---

STEP 1 — ESTABLISH THE ARCHITECTURE

Before writing any KUD content, establish and state:

1a. Knowledge type classification. If {{knowledge_type}} is supplied, use it. If not, classify now:

• T1 (Hierarchical): content builds on hard prerequisites; missing earlier content makes later content inaccessible, not merely harder; canonical correct answers exist. Assessment: rubric with clear criteria.
• T2 (Horizontal): analytical capability deployed when a situation calls for it; better and worse reasoning but no single correct answer; can be applied in multiple domains. Assessment: reasoning-quality rubric on novel scenarios.
• T3 (Dispositional): capacity that only exists as enacted over time and across multiple contexts; no single task is adequate evidence; the pattern of behaviour IS the capability. Assessment: multi-informant observation, not summative rubric.

State the classification and one sentence of reasoning.

1b. Do evidence type. For this LT, the Do cells will be:

PERFORMANCE — if the student produces or executes a discrete evaluable product or act (essay, presentation, seminar contribution, lab report, scenario response, design, constructed artefact). Assessed by rubric on a single occasion. Do language: "I can [verb] [product or performance type]."

DISPOSITION — if the student demonstrates a pattern of behaviour across time and contexts that cannot be adequately evidenced in a single occasion. Assessed through multi-informant observation. Do language: third-person observation indicator ("The student consistently [observable behaviour]" / "The teacher notices [specific pattern]").

Note: most T1 and T2 LTs have performance Dos. Most T3 LTs have dispositional Dos. Exception: some T3 capabilities have a performed expression assessed through multiple rubric occasions over time — T3 means slow development across bands, not automatically observation-only. If this exception applies, state it explicitly.

1c. Know placement. Read {{competency_context}}. If Know content lives in a designated knowledge competency by design (e.g. a Wellbeing Science competency that holds all factual grounding), Know cells in this LT should cross-reference that competency rather than duplicate its content. Write "Know content for this band lives in [LT reference]" rather than restating the content.

1d. Band scope. Note any bands marked N/A with one-sentence developmental rationale.

---

STEP 2 — AUTHOR THE KUD CHART

For each band in scope, produce Know, Understand, Do, Progression lever, and Disciplinary warrant.

KNOW

Specific, testable factual content the student needs to hold.

Each Know item must be a specific fact, not a topic or theme. The test: could you write an exam question directly from this item? If not, decompose it.

WEAK (topic): "Relationships between sleep, food, and movement and daily feeling-states" STRONG (fact): "Sleep deprivation reduces prefrontal cortex activity, which impairs both decision-making and emotional regulation" (Walker, M. (2017). Why We Sleep. Scribner.)

WEAK (topic): "How algorithms shape what we see online" STRONG (fact): "Social media recommendation algorithms are optimised for engagement, not accuracy or wellbeing — content that provokes strong emotional reactions (outrage, anxiety, desire) is amplified because it generates more clicks" (Pariser, E. (2011). The Filter Bubble. Penguin.)

When you encounter a topic-level Know item, decompose it into 2-4 specific facts rather than flagging it. Fix during authoring, not after.

If Know lives in a designated knowledge competency by design: write the cross-reference, not the content.

For T3 dispositional LTs where Know exists only to enrich the disposition: keep the Know layer to 1-2 specific items per band maximum. Do not invent Know content to fill a cell.

UNDERSTAND

One transferable idea per band that a student carries into novel situations they have not encountered before.

An Understand is not a fact. Not a restatement of Do. Not a summary of the lesson. It is the generative insight underneath the capability — the kind of thing a student might say three years later in a completely different context and you think "yes, that is exactly what we were trying to give them."

Independence test: Could a student read the Do statement in the same cell and derive this Understand without additional thinking? If yes, the Understand is not independent. Rewrite it.

Transfer test: Would a student who genuinely held this idea behave differently in a situation they have never encountered before? If not, it is too thin. Rewrite it.

Write as a direct claim or first-person insight. Not a definition.

Examples from multiple domains:

WEAK: "How I talk to people affects how they feel." (restates the Do; no transfer value) STRONG: "What I feel is often set off by something specific, and I can learn what sets me off." (Gross, J. J. (2002). Emotion regulation: Affective, cognitive, and social consequences. Psychophysiology.)

WEAK: "Sources can be biased." (fact, not a transferable insight) STRONG: "Even credible sources have purposes — knowing who wrote something and why changes how I read it." (Wineburg, S. (2021). Why Learn History (When It's Already on Your Phone). University of Chicago Press.)

WEAK: "Physical activity is good for mental health." (too thin — no novel situation transfer) STRONG: "My body and my mood are the same system — what I do with one always affects the other, whether I notice it or not." (Ratey, J. J. (2008). Spark: The Revolutionary New Science of Exercise and the Brain. Little, Brown.)

Understands should become richer and more precise across bands — not just covering harder content but deepening the same insight. Band A Understand: simple direct claim. Band F Understand: nuanced, contextualised, aware of its own limits.

Include the disciplinary warrant per Understand: Author, A. A. (Year). Title. Publisher. One citation per Understand.

DO

Match the evidence type established in Step 1b.

Performance Do: "I can [active verb] [specific product or performance type]."

• Name what is produced or performed specifically. Not "I can demonstrate understanding." What is the artefact or act?
• Examples: "I can write a 300-word analytical argument identifying the main claim, supporting reasoning, and one counterargument." / "I can present a 3-minute structured explanation of a chosen topic using at least two pieces of evidence." / "I can complete a scenario analysis applying the PROVE routine to a provided article."

Disposition Do: "The student consistently [specific observable behaviour in specific contexts]." or "The teacher notices [specific pattern across occasions]."

• Third person. Observable. Specific to this band's developmental level.
• Examples: "The student notices what their body is telling them and names the feeling before responding, without adult prompting." / "The teacher notices the student seeking out perspectives different from their own before forming a position." / "The student acknowledges their part in a conflict and proposes a repair step, without being asked."

Do not write: "I can understand…" / "I can know…" / "I can appreciate…" — these are not observable. Do not write vague dispositional language for T1/T2 LTs.

PROGRESSION LEVER

Name the operative lever(s) at each band transition. Choose from: independence, complexity, scope, precision, reasoning, transfer.

Explain in one phrase what changes: e.g. "Complexity: single trigger → multiple triggers" or "Transfer: practised contexts → novel situations under stress."

If the primary difference between two bands is harder content (topic escalation), this is a progression failure. Rewrite the band statements so the progression is in what students can do with content, not what content they cover.

DISCIPLINARY WARRANT

One citation per Understand. Format: Author, A. A. (Year). Title. Publisher/Journal. This is the minimum — "informed educator level" is not sufficient. A teacher should be able to follow up the citation.

---

STEP 3 — PREREQUISITE TYPING

For each prerequisite relationship identified (from {{prerequisite_lts}} or discovered during authoring), type it:

HARD — understanding B is logically impossible without A. The knowledge does not merely help — its absence makes the target content inaccessible. Warrants a readiness check before beginning. Example: evaluating health claims (T1) requires a working model of how the stress-response system operates — without the factual model, "evaluation" is guesswork.

SOFT ENABLER — understanding A enriches, accelerates, or deepens B, but B is reachable through other pathways including experiential and embodied ones. Do not gate assessment of the dependent LT because the enabler is not yet in place. Example: neuroscience of stress enriches self-regulation but a student can develop genuine self-regulation through practised strategies and adult co-regulation without understanding the amygdala.

CONCEPTUAL ACCELERATOR — A dramatically improves the portability and transferability of B to novel contexts. It teaches why the competency works, making it recognisable in new situations and teachable to others. Example: understanding conditioning mechanisms makes digital wellbeing strategies portable to any new platform or context, not just the ones practised in class.

Most knowledge-to-disposition relationships are soft enablers or conceptual accelerators. Do not designate a relationship as hard unless the logical dependency is genuine.

Record prerequisites in two places:

1. LT header: name the prerequisite LT by ID and type. e.g. "LT 6.1 — soft enabler (conceptual accelerator)"
2. KUD band level: flag where a specific K or U item at a specific band depends on another LT's content. e.g. "Band C U-layer: conceptual accelerator — LT 6.1 Band C in place makes this Understand portable; proceed without it if LT 6.1 Band C is not yet established."

---

STEP 4 — QUALITY CHECKS

Run all six checks before producing final output. For each check, report PASS or FLAG. If FLAG: name the specific cell (e.g. "Band C, Understand") and state exactly what needs to change.

Check 1 — Understand independence: Is each Understand in a cell genuinely distinct from the Do in the same cell? Could a student read only the Do statement and derive the Understand without additional thinking? If yes: FLAG. Rewrite the Understand so it adds something the Do does not contain.

Check 2 — Understand transfer: Would a student who genuinely held each Understand behave differently in a novel situation they have not encountered before? If the Understand only applies in the context where it was learned: FLAG. Rewrite for genuine transfer.

Check 3 — Know specificity: Is each Know item a specific testable fact? Could you write a direct exam question from it? If it is a topic or theme: FLAG and decompose it into 2-4 specific facts in your output. Fix during this check — do not leave topic-level items in the output.

Check 4 — Progression lever: Is each band-to-band transition explicable by a named lever? If the only change between bands is harder content (more complex topics, higher reading level, more examples): FLAG as topic escalation. Rewrite so the progression is in what students can do with content.

Check 5 — Compound knowledge-type: Does this LT bundle T2 assessable analytical content with T3 dispositional content? The test: would the Do performance or observation adequately evidence the K/U content, or does the K/U content require a separate task format to be evidenced? Example: if the Do is a Socratic seminar contribution and the K/U content is factual recall of specific mechanisms, the seminar contribution does not adequately evidence the recall — FLAG for split. If the K/U content is enrichment that informs the disposition without requiring separate assessment — PASS.

Check 6 — Know placement: For LTs where Know lives in a designated knowledge competency by design — does any Know cell duplicate that competency's content rather than cross-reference it? If yes: FLAG and replace with the cross-reference.

---

STEP 5 — INPUTS

• input_type: {{input_type}}
• primary_input: {{primary_input}}
• lt_name: {{lt_name}}
• lt_definition: {{lt_definition}}
• knowledge_type: {{knowledge_type}}
• band_scope: {{band_scope}}
• competency_context: {{competency_context}}
• crosswalk_data: {{crosswalk_data}}
• prerequisite_lts: {{prerequisite_lts}}
• programme_purpose: {{programme_purpose}}
• mode: {{mode}}

---

SCHOOL-AGNOSTIC OUTPUTS

SCHOOL-AGNOSTIC OUTPUTS: Generated KUD charts must not reference any school-specific units, programmes, curriculum documents, or named courses. The skill works from the LT definition, band theory, and knowledge type classification only. If the operator has provided a school programme document as context, that document is background information for the operator's use — its content, unit names, and programme structures must not appear anywhere in the generated output. Any output referencing a named school unit or programme is invalid and must be regenerated without that reference.

---

OUTPUT FORMAT

Return output in this structure:

Routing note: [null or recommendation text — always present]

---

KUD Chart: [LT Name]

LT Definition: [one sentence, "I can…"] Knowledge Type: [T1 / T2 / T3 — with one sentence reasoning] Do Evidence Type: [Performance / Disposition / Mixed — with note if T3 exception applies] Band Scope: [A–F or subset with N/A rationale] Prerequisites: [LT ID — type (hard / soft enabler / conceptual accelerator) — one sentence rationale; or "None identified"]

Band	Know	Understand	Do	Progression lever	Disciplinary warrant
A
B
C
D
E
F

---

Quality Check Results

Check 1 — Understand independence: [PASS / FLAG: cell + fix required] Check 2 — Understand transfer: [PASS / FLAG: cell + fix required] Check 3 — Know specificity: [PASS / FLAG: cell + decomposed facts] Check 4 — Progression lever: [PASS / FLAG: cell + rewrite] Check 5 — Compound knowledge-type: [PASS / FLAG: split test result] Check 6 — Know placement: [PASS / FLAG: cell + cross-reference]

---

Authoring Notes

[Decisions made during authoring requiring teacher review: ambiguous classification calls, cells where the warrant is thin, bands where progression logic is uncertain, any deviation from the brief.]

Example Output

Input: LT 1.3 — Personal Identity & Cultural Self-Awareness. input_type: existing_lt_with_do. Knowledge type: T3. Competency: C1 — Emotional Intelligence. Know lives in C6 by design. Band scope: A–F. Programme purpose: REAL School Budapest — develop learners who understand themselves across cultural contexts and navigate identity in international settings.

Routing note: null

---

KUD Chart: LT 1.3 — Personal Identity & Cultural Self-Awareness

LT Definition: I can articulate a developing sense of who I am — my values, strengths, and cultural influences — and remain grounded in that identity across different contexts and communities. Knowledge Type: T3 — Dispositional. Identity formation develops through sustained experience across contexts; no single occasion adequately evidences the capacity. Do Evidence Type: Disposition. Observable pattern across time — the teacher watches for stability, curiosity about self, and grounded engagement across settings. Band Scope: A–F Prerequisites: LT 1.1 — soft enabler (self-awareness of emotional states is the developmental precursor to identity reflection; does not gate assessment)

Band	Know	Understand	Do	Progression lever	Disciplinary warrant
A	Know content lives in LT 6.1	I am someone — I have things I like, things I'm good at, and people who matter to me.	The student can name three things they like about themselves and describe what makes them different from and similar to a classmate.	Baseline	Wellman, H. M. (2002). Understanding the psychological world. In U. Goswami (Ed.), Blackwell Handbook of Childhood Cognitive Development. Blackwell.
B	Know content lives in LT 6.1	The same me shows up differently in different places — I act differently at home, at school, and with friends, but something stays the same.	The teacher notices the student referring to personal interests or values when making choices, without prompting. The student can describe one way they are the same and one way they are different across two settings.	Complexity: single-context self → multi-context self	Harter, S. (1999). The Construction of the Self. Guilford Press.
C	Know content lives in LT 6.1	My background and the cultures I belong to shape what I notice and value — and other people's backgrounds shape what they notice and value.	The teacher notices the student showing curiosity about how others' cultural backgrounds shape their perspectives, asking questions rather than assuming.	Scope: self-understanding → self-in-relation-to-others	Markus, H. R. & Kitayama, S. (1991). Culture and the self: Implications for cognition, emotion, and motivation. Psychological Review, 98(2), 224–253.
D	Know content lives in LT 6.1	My identity is not fixed — it is something I am actively building, and I have more agency over it than I sometimes think.	The teacher notices the student reflecting on how their sense of self has changed and naming specific experiences that contributed to that change, without prompting.	Reasoning: noticing → analysing identity development	Marcia, J. E. (1966). Development and validation of ego-identity status. Journal of Personality and Social Psychology, 3(5), 551–558.
E	Know content lives in LT 6.1	When cultural expectations conflict with my own values, the tension is real — and navigating it honestly, rather than suppressing one side, is how identity becomes coherent.	The teacher notices the student naming cultural pressures they feel and articulating how they are deciding what to keep and what to question, without needing external validation.	Reasoning: analysis → principled navigation of conflict	Berry, J. W. (2005). Acculturation: Living successfully in two cultures. International Journal of Intercultural Relations, 29(6), 697–712.
F	Know content lives in LT 6.1	Who I am is partly my own construction and partly shaped by forces I did not choose — and holding both of those truths at once is what it means to have a mature sense of self.	The teacher notices the student articulating a coherent sense of self that integrates personal history, cultural background, and chosen values — and applying this coherence when facing genuinely novel situations.	Transfer: known contexts → novel situations; Reasoning: analysis → integration	Kegan, R. (1994). In Over Our Heads: The Mental Demands of Modern Life. Harvard University Press.

---

Quality Check Results

Check 1 — Understand independence: PASS — each Understand adds something not derivable from the Do alone. Check 2 — Understand transfer: PASS — each Understand would change behaviour in novel cultural or social situations. Check 3 — Know specificity: PASS — all Know cells cross-reference C6 by design; no topic-level items present. Check 4 — Progression lever: PASS — baseline → complexity → scope → reasoning → reasoning → transfer. No topic escalation. Check 5 — Compound knowledge-type: PASS — K/U content enriches the dispositional Do; no separate rubric task would be given for K/U content in this LT. Check 6 — Know placement: PASS — all Know cells cross-reference LT 6.1 rather than duplicating.

---

Authoring Notes

Band A Do was written as a performed element (naming three things, describing similarities/differences) rather than a pure observation indicator because at Band A the disposition is too nascent to observe reliably across contexts — a structured activity produces more valid evidence at this developmental stage. Teacher should review whether this matches their assessment practice. Band F Understand draws on Kegan's constructive-developmental theory — teacher should verify this aligns with the programme's philosophical stance on self-authorship before adopting.

Known Limitations

The Understand quality checks are the hardest to pass reliably. Whether an Understand is genuinely transferable or merely paraphrases the Do requires judgement about what counts as a novel situation — the skill's assessment should always be reviewed by the teacher before adopting. Teachers who have taught the capability for several years will have better intuitions about what Understands actually transfer than any automated check.

The routing check produces a note but never blocks execution. This means a skill run on a complex raw curriculum document will produce a draft with weaker Know specificity and progression coherence than a run on structured harness output. The authoring notes will flag this, but the teacher should not treat a direct-from-document draft as production-ready without additional review.

Prerequisite typing is based on the logical structure of the content as the skill can assess it from the inputs supplied. The teacher who knows the student population and has observed prerequisite gaps in practice will have more accurate information than the skill about whether a given relationship is functioning as a hard gate or a soft enabler in their specific context.

The performance vs disposition distinction for some T3 LTs with performed expressions (e.g. presentation skill developed over multiple occasions) requires teacher judgement about whether the capability being developed is primarily a skill that can be evidenced in a single well-prepared performance, or a disposition that requires accumulated evidence across unprepared real-world occasions. The skill will flag this ambiguity in authoring notes when it arises.

Assessment Validity Checker

What This Skill Does

Evaluates a proposed assessment against three dimensions: validity (does it measure what it claims to measure?), reliability (would different markers agree on the score?), and authenticity (is the task meaningful and does it require genuine demonstration of the intended learning?). The output identifies specific threats to validity — construct-irrelevant variance (the assessment measures something other than what it claims), construct underrepresentation (the assessment doesn't cover enough of what it claims to measure), and consequential validity problems (unintended negative effects of the assessment) — and provides specific, actionable recommendations for each threat. AI is specifically valuable here because most teacher-designed assessments contain validity threats that are invisible without explicit analytical frameworks — a teacher designing a "reading comprehension" test may inadvertently create a writing test, or a "science understanding" assessment may actually test literacy.

Evidence Foundation

Messick (1989) unified the concept of validity into a single framework: validity is not a property of a test but of the interpretation and use of test scores. A test is not "valid" or "invalid" in the abstract — it is valid FOR a specific purpose with a specific population. This means every assessment must be evaluated against its intended use. Wiliam (2011) applied this framework to classroom assessment, showing that the most common validity threat in teacher-designed assessment is construct-irrelevant variance — where the assessment measures something other than the intended construct. For example, a group presentation assessed for "understanding of climate change" may actually measure public speaking confidence, group dynamics, and technology skills more than climate change understanding. Kane (2006) proposed a validation-as-argument approach: the validity of an assessment depends on the strength of the chain of reasoning from the task → the response → the score → the interpretation → the decision. Any weak link in this chain is a validity threat. Brookhart (2003) adapted measurement theory for classroom contexts, arguing that classroom assessments need not meet the same psychometric standards as standardised tests but must still demonstrate that they measure what they claim. Stobart (2008) highlighted consequential validity — the effects of assessment on learning. If an assessment drives students toward surface learning, test anxiety, or strategic behaviour rather than genuine engagement, its consequential validity is compromised.

Input Schema

The teacher must provide:

• Assessment description: What students do and how it's marked. e.g. "Students write a 500-word essay on the causes of WW1, marked against a rubric with four criteria: historical knowledge, analytical argument, use of evidence, and written communication" / "Students complete a 30-question multiple choice test on photosynthesis" / "Students create a poster about healthy eating and present it to the class"
• Intended learning: What the assessment claims to measure. e.g. "Understanding of the causes of WW1 and ability to construct a historical argument" / "Knowledge and understanding of photosynthesis" / "Understanding of nutrition and healthy eating"
• Student level: Year group. e.g. "Year 10"

Optional (injected by context engine if available):

• Subject area: The curriculum subject
• Assessment purpose: Formative, summative, diagnostic, or evaluative
• Marking approach: How it will be marked
• Stakes: Consequences of the assessment

Prompt

You are an expert in educational assessment and measurement, with deep knowledge of Messick's (1989) unified validity framework, Wiliam's (2011) approach to classroom assessment validity, Kane's (2006) argument-based validation, and Stobart's (2008) work on consequential validity. You understand that validity is not a property of the test itself but of the interpretation and use of the scores — an assessment is valid FOR a specific purpose, and the same assessment may be valid for one purpose but invalid for another.

Your task is to evaluate the validity of:

**Assessment description:** {{assessment_description}}
**Intended learning:** {{intended_learning}}
**Student level:** {{student_level}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Subject area:** {{subject_area}} — if not provided, infer from the assessment description.
**Assessment purpose:** {{assessment_purpose}} — if not provided, infer from the description and stakes.
**Marking approach:** {{marking_approach}} — if not provided, note that marking approach affects reliability and recommend one.
**Stakes:** {{stakes}} — if not provided, analyse for both low-stakes and high-stakes use.

Analyse across these dimensions:

1. **Construct validity (Messick, 1989):**
   - Does the assessment task actually require demonstration of the intended learning?
   - **Construct-irrelevant variance:** Does the assessment inadvertently measure something ELSE? (E.g., a science assessment that requires essay writing also measures literacy; a group project also measures collaboration and social dynamics.)
   - **Construct underrepresentation:** Does the assessment cover enough of the intended construct? (E.g., a test on "understanding photosynthesis" that only asks factual recall questions doesn't assess understanding — it assesses memorisation.)

2. **Content validity:**
   - Does the assessment sample appropriately from the domain of intended learning?
   - Are important aspects of the learning missing from the assessment?
   - Are there aspects of the assessment that go beyond the intended learning?

3. **Reliability (Brookhart, 2003):**
   - Would different markers give the same score? (Inter-rater reliability)
   - Would the same student get a similar score on a different day? (Test-retest)
   - Is the marking scheme clear enough to be applied consistently?

4. **Consequential validity (Stobart, 2008):**
   - What behaviours will this assessment drive? Will students engage in genuine learning or strategic/surface-level preparation?
   - Does the assessment create unfair barriers for specific groups (EAL students, students with learning differences)?
   - Does the assessment format match the learning it claims to measure?

5. **Authenticity:**
   - Does the task require genuine demonstration of the learning, or can a student perform well without actually having learned the intended content?
   - Can the task be completed through memorisation, copying, or procedural compliance without understanding?

Return your output in this exact format:

## Assessment Validity Analysis

**Assessment:** [Brief description]
**Claims to measure:** [Intended learning]
**For:** [Student level]

### Validity Analysis

**Construct validity:** [Analysis of whether the assessment measures what it claims]
**Construct-irrelevant variance:** [What ELSE the assessment measures that isn't intended]
**Construct underrepresentation:** [What aspects of the intended learning are NOT assessed]

### Reliability Analysis

[Analysis of marking consistency and specific reliability concerns]

### Consequential Validity

[What learning behaviours the assessment is likely to drive]

### Threats Identified

[Numbered list of specific validity threats, each with an explanation]

### Recommendations

[Specific, actionable modifications to address each identified threat]

### Overall Verdict

[Summary judgement: is this assessment fit for its stated purpose? What is the single most important modification?]

**Self-check before returning output:** Verify that (a) the analysis addresses construct validity, reliability, and consequential validity, (b) each threat is specific and explained, (c) each recommendation directly addresses an identified threat, (d) the analysis considers the assessment's purpose and stakes, and (e) the verdict is constructive — identifying what to fix, not just what's wrong.

Example Output

Scenario: Assessment description: "Students create a poster about the causes of climate change and present it to the class. The poster must include at least 3 causes, a visual element (graph or diagram), and be 'visually appealing.' Marked out of 20: content (8 marks), visual design (6 marks), presentation (6 marks)." / Intended learning: "Students understand the causes and effects of climate change." / Student level: "Year 9" / Subject area: "Geography"

---

Assessment Validity Analysis

Assessment: Poster presentation on climate change causes Claims to measure: Understanding of the causes and effects of climate change For: Year 9 Geography

Validity Analysis

Construct validity: MODERATE CONCERN. The assessment claims to measure understanding of climate change causes and effects, but the task and marking scheme assess three constructs simultaneously: geography knowledge (8/20 = 40%), visual design skill (6/20 = 30%), and public speaking skill (6/20 = 30%). A student who deeply understands climate change but cannot draw or speak confidently will score lower than a student with superficial knowledge who presents a beautifully designed, confidently delivered poster. The assessment does not primarily measure what it claims to.

Construct-irrelevant variance:

• Visual design (6 marks): The ability to create a "visually appealing" poster is unrelated to understanding climate change. This measures art/design skills and access to coloured pens, printers, and craft materials — resources that correlate with socioeconomic background, not geographic understanding. A student who produces an aesthetically poor but scientifically accurate poster is penalised. This is the most significant validity threat.
• Presentation skills (6 marks): Public speaking confidence is a separate skill from geographic understanding. EAL students, students with speech difficulties, and students with social anxiety are disadvantaged by this component. The presentation may reveal understanding, but 6/20 marks for delivery means presentation manner is weighted as heavily as scientific accuracy.
• Material access: Students who can print colour images, use computers at home, or purchase poster materials have an advantage unrelated to learning.

Construct underrepresentation:

• The assessment requires "at least 3 causes" but the intended learning includes "effects" — effects are not assessed despite being part of the stated learning objective.
• The assessment doesn't require students to EXPLAIN causes — only to list them. A poster listing "burning fossil fuels, deforestation, agriculture" without explanation can score full content marks. Listing is recall; explaining is understanding. The assessment underrepresents the "understanding" component.
• No assessment of relationships between causes, relative significance, or evidence evaluation — which are the geographic thinking skills the objective implies.

Reliability Analysis

• "Visually appealing" is subjective and unanchored. Two markers are likely to disagree significantly on what counts as visually appealing. This criterion will produce the lowest inter-rater reliability of any component.
• Content marking (8 marks) could be reliable if the mark scheme specifies what counts as a "cause" and what level of explanation earns each mark — but as described, "at least 3 causes" is a binary threshold, not a quality scale. A student listing 3 basic causes and a student providing detailed, evidenced explanations of 5 causes may receive similar scores.
• Presentation (6 marks) has no described criteria. Markers may weight fluency, confidence, eye contact, volume, or content accuracy differently. Without a rubric, reliability will be low.

Consequential Validity

• The assessment incentivises decoration over understanding. Students who recognise that 12/20 marks (60%) depend on design and presentation will rationally invest more time on visuals and rehearsal than on learning the content. This drives surface-level learning behaviour.
• The assessment disadvantages EAL students and shy students. 6 marks for presentation creates a barrier that is unrelated to geography. Students who understand climate change deeply but are anxious about presenting will underperform relative to their actual understanding.
• The assessment rewards socioeconomic advantage. Access to printers, coloured paper, markers, and a computer for design gives some students a significant edge on 30% of the marks. This is inequitable.

Threats Identified

1. MAJOR: Construct-irrelevant variance from visual design. 30% of marks assess design skill, not geography. This means the assessment score does not represent what it claims.
2. MAJOR: Construct-irrelevant variance from presentation. 30% of marks assess public speaking, disadvantaging EAL, anxious, and speech-difficulty students.
3. MODERATE: Construct underrepresentation. Effects of climate change are not assessed despite being in the intended learning. Explanation of causes is not required — listing is sufficient.
4. MODERATE: Low reliability. "Visually appealing" is subjective. Presentation marks lack a rubric. Content marks don't distinguish quality of explanation.
5. MINOR: Consequential validity. Mark weighting incentivises decoration over learning.

Recommendations

1. Restructure the mark allocation. Content should be at least 70% of the total marks. Reduce visual design to 0 marks (remove it as a criterion entirely) or cap at 2 marks for "clarity of visual communication" (whether the diagram/graph communicates information effectively — not whether it looks pretty). Reduce presentation to 4 marks maximum, with a rubric focused on clarity of explanation, not confidence of delivery.

2. Require explanation, not just listing. Change "at least 3 causes" to "explain 3 causes of climate change, including HOW each cause contributes to increased greenhouse gas concentrations." This requires understanding, not recall.

3. Add the missing construct. Add a requirement to address effects as well as causes — or change the intended learning to match the assessment. Alignment must go both ways.

4. Replace "visually appealing" with "clear and communicative." If visual elements remain, assess whether the graph/diagram communicates geographic information accurately — not whether it looks attractive.

5. Provide an alternative to live presentation. Allow students to submit a recorded explanation, a written paragraph, or a one-to-one explanation to the teacher instead of a class presentation. This removes the public speaking barrier while still assessing whether the student can explain their understanding.

6. Create a content rubric. Replace the 8-mark content score with a descriptive rubric that distinguishes between listing causes (lower level), explaining causes (middle level), and analysing the relationships and relative significance of causes (higher level).

Overall Verdict

As currently designed, this assessment is not fit for purpose. It claims to measure understanding of climate change but actually measures a blend of geography knowledge (40%), art skills (30%), and public speaking (30%). The single most important modification is restructuring the mark allocation so that content understanding dominates the assessment. If a student deeply understands climate change causes and effects, they should score highly — regardless of their poster design or presentation confidence.

---

Known Limitations

1. The analysis evaluates the assessment as described, which may differ from how it's implemented. A teacher who marks generously on design and strictly on content may partially compensate for the mark allocation issue in practice — but the structural problem remains. The assessment's design, not just its implementation, should be valid.

2. Validity is always relative to purpose. This analysis evaluates validity for the STATED purpose (measuring understanding of climate change). If the assessment's actual purpose includes developing presentation skills, the validity analysis would differ — but the assessment should then be labelled as measuring multiple constructs.

3. Some validity threats are trade-offs, not errors. Including a presentation component may have legitimate pedagogical reasons (building oracy skills, developing confidence). The analysis identifies the validity cost of these design choices — the teacher must decide whether the pedagogical benefits justify the validity compromise. The key is being transparent about what the assessment actually measures.

Backwards Design Unit Planner

What This Skill Does

Generates a complete Stage 1–2–3 Understanding by Design unit structure from a teacher's desired outcomes: Stage 1 defines enduring understandings, essential questions, and target knowledge/skills; Stage 2 designs the assessment evidence that will demonstrate understanding (before any activities are planned); Stage 3 sequences the learning activities that build toward the assessments and outcomes. The critical insight of backwards design is that assessment is designed BEFORE instruction — not as an afterthought but as the definition of what success looks like. AI is specifically valuable here because backwards design requires holding all three stages in mind simultaneously and ensuring tight alignment between them — what is assessed must match what is intended, and what is taught must prepare students for what is assessed. Most teacher-designed units plan activities first and assessments last, which produces misalignment.

Evidence Foundation

Wiggins & McTighe (1998, 2005) developed Understanding by Design (UbD), the most widely adopted curriculum design framework in education. The framework's central argument is that curriculum should be designed backward from desired results, not forward from available activities. Stage 1 (Desired Results) defines what students should understand — not just know or do, but genuinely understand at a transferable level. Stage 2 (Assessment Evidence) determines what evidence would demonstrate that understanding — designed before instruction so that teaching targets real outcomes, not just coverage. Stage 3 (Learning Plan) sequences the instruction needed to build toward the assessed outcomes. Wiggins & McTighe (2011) provided practical guidance for unit creation, emphasising that enduring understandings should be transferable ideas worth understanding beyond the unit, and essential questions should be genuinely open — questions that provoke inquiry rather than have predetermined answers. Biggs & Tang (2011) developed "constructive alignment" — the principle that learning outcomes, assessment tasks, and teaching activities must be aligned so that what is assessed is what is taught and what is taught prepares for what is assessed. Hattie (2009) confirmed that clarity of learning intentions and success criteria is one of the highest-leverage factors in student achievement.

Input Schema

The teacher must provide:

• Desired outcomes: What students should understand, know, and be able to do. e.g. "Students will understand how organisms adapt to their environments through natural selection, know key vocabulary (adaptation, variation, natural selection, evolution), and be able to explain how a specific organism's traits relate to its environment."
• Student level: Year group. e.g. "Year 9"
• Unit duration: Length of the unit. e.g. "6 lessons (1 hour each)" / "3 weeks"

Optional (injected by context engine if available):

• Subject area: The curriculum subject
• Curriculum framework: Specific standards
• Student profiles: Prior attainment, known gaps
• Available resources: Texts, materials, resources

Prompt

You are an expert in curriculum design, with deep knowledge of Wiggins & McTighe's (1998, 2005) Understanding by Design framework, Biggs & Tang's (2011) constructive alignment, and Hattie's (2009) research on learning intentions and success criteria. You understand that effective unit design works BACKWARD from desired results — defining what students will understand first, then designing assessment evidence, then planning learning activities.

Your task is to design a UbD unit for:

**Desired outcomes:** {{desired_outcomes}}
**Student level:** {{student_level}}
**Unit duration:** {{unit_duration}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Subject area:** {{subject_area}} — if not provided, infer from the desired outcomes.
**Curriculum framework:** {{curriculum_framework}} — if not provided, design in general terms.
**Student profiles:** {{student_profiles}} — if not provided, assume a typical mixed-ability class.
**Available resources:** {{available_resources}} — if not provided, suggest appropriate resources.

Apply these evidence-based principles:

1. **Stage 1 — Desired Results (Wiggins & McTighe, 2005):**
   - **Enduring understandings:** Big ideas that are transferable beyond this unit. These are not facts to memorise but principles to understand. Frame as "Students will understand that..." statements.
   - **Essential questions:** Open, thought-provoking questions that guide inquiry throughout the unit. These should be genuinely debatable — not questions with a single right answer. They should recur throughout the unit, with students' answers deepening over time.
   - **Knowledge:** Specific facts, concepts, and vocabulary students will know.
   - **Skills:** Specific abilities students will be able to demonstrate.

2. **Stage 2 — Assessment Evidence (Wiggins & McTighe, 2005; Biggs & Tang, 2011):**
   - Design assessment BEFORE instruction. This is the core UbD principle.
   - **Performance task:** A rich, authentic task that requires students to demonstrate understanding through application — not just recall facts. The task should require transfer — applying learning to a new situation.
   - **Other evidence:** Additional assessment methods (quizzes, checks for understanding, observations) that gather evidence of knowledge and skills.
   - Assessment must be aligned to Stage 1 — every enduring understanding and essential question must be assessable through the evidence in Stage 2.

3. **Stage 3 — Learning Plan (Wiggins & McTighe, 2005):**
   - Sequence learning activities that build toward the Stage 2 assessments.
   - Use the WHERETO framework: Where are we going? Hook the student. Explore and equip. Rethink and revise. Evaluate. Tailor to individual needs. Organise for understanding.
   - Activities should be sequenced logically — building knowledge before applying it, scaffolding before independence.
   - Each activity should connect clearly to the Stage 2 assessment — if an activity doesn't prepare students for the assessment, question whether it belongs.

4. **Alignment check (Biggs & Tang, 2011):**
   - Verify that Stage 1 outcomes are assessed in Stage 2 and taught in Stage 3.
   - Flag any misalignment: outcomes that are stated but not assessed, or assessed but not taught.

Return your output in this exact format:

## Unit Plan: [Unit Title]

**For:** [Student level]
**Subject:** [Subject area]
**Duration:** [Unit duration]

### Stage 1: Desired Results

**Enduring Understandings**
[2–3 transferable understandings — "Students will understand that..."]

**Essential Questions**
[2–3 open, recurring questions that guide inquiry]

**Students will know:**
[Specific knowledge — facts, concepts, vocabulary]

**Students will be able to:**
[Specific skills]

### Stage 2: Assessment Evidence

**Performance Task**
[Description of the authentic assessment task, including: the scenario, what students produce, what the task assesses, and brief success criteria]

**Other Evidence**
[Additional assessment methods mapped to specific Stage 1 outcomes]

### Stage 3: Learning Plan

[Lesson-by-lesson or phase-by-phase sequence, showing: what happens, what it builds toward, and how it connects to Stage 2]

### Alignment Check

[Verification that all Stage 1 outcomes are assessed in Stage 2 and addressed in Stage 3. Flag any gaps.]

**Self-check before returning output:** Verify that (a) enduring understandings are transferable ideas, not just facts, (b) essential questions are genuinely open and recurring, (c) the performance task requires application/transfer, not just recall, (d) Stage 2 was designed before Stage 3 in logic (assessment determines instruction), (e) every Stage 1 outcome is assessed and taught, and (f) the learning plan builds progressively toward the performance task.

Example Output

Scenario: Desired outcomes: "Students will understand how natural selection drives adaptation in organisms, know key concepts (adaptation, variation, inheritance, natural selection, evolution), and be able to explain how specific organisms' traits are suited to their environment and how populations change over time." / Student level: "Year 9" / Unit duration: "8 lessons (50 minutes each)" / Subject area: "Science (Biology)"

---

Unit Plan: Natural Selection and Adaptation

For: Year 9 Subject: Science (Biology) Duration: 8 lessons (50 minutes each)

Stage 1: Desired Results

Enduring Understandings

1. Students will understand that organisms with traits better suited to their environment are more likely to survive and reproduce, and these traits are passed to offspring — leading to populations that change over time. (This is the mechanism of natural selection.)
2. Students will understand that variation within a population is essential for natural selection — without variation, there is nothing for the environment to "select."
3. Students will understand that adaptation is a result of natural selection over many generations, not a deliberate choice by individual organisms. (Organisms do not choose to adapt; populations change because better-suited individuals leave more offspring.)

Essential Questions

1. "Why do organisms look the way they do?" (Recurring — answers deepen from "because they adapted" in Lesson 1 to "because natural selection favoured traits that increased survival and reproduction in their specific environment" by Lesson 8.)
2. "If the environment changes, what happens to organisms that are well-adapted to the old environment?" (Drives the understanding that adaptation is not permanent — it is context-dependent.)
3. "Why is variation important?" (Connects genetics to evolution — without variation, natural selection cannot operate.)

Students will know:

• The definitions of adaptation, variation, inheritance, natural selection, evolution, and species
• Examples of structural, behavioural, and functional adaptations in named organisms
• That variation arises from genetic differences (and sometimes environmental factors)
• The four conditions required for natural selection: variation, inheritance, selection pressure, differential reproduction
• Darwin's contribution to the theory of evolution by natural selection

Students will be able to:

• Explain how a named organism's specific traits help it survive in its environment
• Describe the process of natural selection step by step
• Predict what might happen to a population if the environment changes
• Distinguish between Lamarckian (acquired characteristics) and Darwinian (natural selection) explanations — and explain why Lamarck's is incorrect

Stage 2: Assessment Evidence

Performance Task: "The Island"

Scenario: "A population of beetles lives on a volcanic island. The beetles vary in colour from green to brown. The island is covered in green vegetation. Recently, a volcanic eruption has covered much of the island in dark grey ash. Scientists predict that the beetle population will look very different in 50 generations. Your task: write a scientific explanation predicting how the beetle population will change and why, using the theory of natural selection. Your explanation must include all four conditions for natural selection and address the common misconception that individual beetles will 'choose' to change colour."

What students produce: A 200–300 word scientific explanation.

What the task assesses:

• Enduring Understanding 1: mechanism of natural selection (can they explain the process correctly?)
• Enduring Understanding 2: role of variation (do they identify existing colour variation as essential?)
• Enduring Understanding 3: adaptation is not a choice (do they explain that the population changes, not individual beetles?)
• Essential Questions 1 and 2: adapted organisms + environmental change
• Knowledge: four conditions, correct terminology
• Skills: step-by-step explanation, distinguishing Darwinian from Lamarckian reasoning

Brief success criteria:

• Identifies existing variation in the population (colour range)
• Explains selection pressure (darker beetles better camouflaged on ash)
• Describes differential survival and reproduction
• Explains inheritance of traits to offspring
• Predicts population change over generations
• Addresses the misconception that beetles "choose" to change colour

Other Evidence

Lesson	Assessment	Stage 1 outcome assessed
2	Exit ticket: "Name 3 adaptations of a polar bear and explain how each helps it survive"	Knowledge (adaptations); Skill (explaining trait–environment link)
4	Mini-whiteboard hinge question: "A giraffe stretches its neck to reach leaves. Its offspring are born with longer necks. Is this Darwinian or Lamarckian? Why?"	Understanding 3 (adaptation is not a choice); Knowledge (Lamarck vs Darwin)
6	Peer explanation: students explain natural selection to a partner using a new example; partner checks against a 4-step checklist	Knowledge (four conditions); Skill (step-by-step explanation)

Stage 3: Learning Plan

Lesson 1 — Hook and Explore: "Why Do Organisms Look the Way They Do?"

• Show 5 images of unusual adaptations (anglerfish, cactus, arctic fox, chameleon, deep-sea tube worm). Students discuss: "Why does this organism look like this? What problem does this trait solve?"
• Introduce Essential Question 1. Students write initial answers.
• Define "adaptation" — structural, behavioural, functional. Students classify the 5 examples.
• Builds toward: Stage 2 performance task (explaining trait–environment link).

Lesson 2 — Equip: Types of Adaptation

• Detailed study of 3 organisms and their adaptations (one structural, one behavioural, one functional).
• Students explain in writing: "How does [trait] help [organism] survive in [environment]?"
• Exit ticket assessment.
• Builds toward: Stage 2 performance task (explaining how traits suit environments).

Lesson 3 — Explore: Variation

• Practical: measuring variation in a population (e.g., hand span, leaf size, or simulated beetle data).
• Key concept: variation exists within every population. Some variation is genetic.
• Introduce Essential Question 3: "Why is variation important?"
• Builds toward: Enduring Understanding 2 (variation is essential for natural selection).

Lesson 4 — Equip: Natural Selection Step by Step

• Explicit instruction: the 4 conditions for natural selection (variation, inheritance, selection pressure, differential reproduction).
• Use the peppered moth as a worked example — model the explanation step by step.
• Introduce Lamarck vs. Darwin distinction. Common misconception: "the giraffe stretched its neck."
• Hinge question assessment.
• Builds toward: Performance task (students must use the 4 conditions).

Lesson 5 — Explore and Rethink: Simulation

• Simulation activity: "Survival of the Fittest" using coloured paper "organisms" on different backgrounds. Students observe which are "eaten" (selected against) and which survive. Repeat for 3 "generations."
• Students observe that the population changes over time — without any individual changing.
• Revisit Essential Question 1: update answers.
• Builds toward: Enduring Understanding 3 (populations change, not individuals).

Lesson 6 — Equip and Evaluate: Practice Explanations

• Students practise writing natural selection explanations using a new example (e.g., antibiotic resistance in bacteria).
• Peer assessment using the 4-step checklist from the performance task criteria.
• Teacher circulates and identifies common errors for whole-class feedback.
• Builds toward: Performance task (practising the exact skill assessed).

Lesson 7 — Performance Task: "The Island"

• Students complete the performance task under supported conditions.
• Sentence starters available for EAL students. Word bank available for all students (may be removed for extension students).
• Teacher circulates, noting common misconceptions for Lesson 8 feedback.

Lesson 8 — Rethink and Extend: Feedback and Transfer

• Return marked work with targeted feedback.
• Whole-class review of most common errors (especially Lamarckian reasoning).
• Extension: "Humans have been using antibiotics for about 80 years. Bacteria are becoming resistant. Use natural selection to explain why."
• Revisit all three Essential Questions: students write final answers and compare with Lesson 1 answers.

Alignment Check

Stage 1 Outcome	Assessed in Stage 2?	Taught in Stage 3?	Aligned?
Understanding 1 (natural selection mechanism)	Performance task + Lesson 6 peer assessment	Lessons 4, 5, 6	✓
Understanding 2 (variation is essential)	Performance task	Lesson 3	✓
Understanding 3 (adaptation is not a choice)	Performance task + Lesson 4 hinge question	Lessons 4, 5	✓
Knowledge (key vocabulary)	All assessments require terminology	Introduced progressively across Lessons 1–4	✓
Skill (explain trait–environment link)	Lesson 2 exit ticket + performance task	Lessons 1, 2	✓
Skill (step-by-step explanation)	Performance task + Lesson 6 peer check	Lessons 4, 6 (modelled and practised)	✓

No misalignment detected. All Stage 1 outcomes are assessed and taught.

---

Known Limitations

1. The unit plan provides structure, not detailed lesson plans. Each lesson entry describes the key activity and its purpose but does not include full timing, differentiation, resources, or teacher scripts. Teachers should use the Stage 3 sequence as a framework and develop detailed lesson plans using other skills (Explicit Instruction Sequence Builder, Lesson Opening Designer, etc.).

2. Backwards design assumes clear desired outcomes. If the teacher's initial outcome statement is vague ("students will learn about natural selection"), the UbD structure will be less precise. The quality of the unit depends on the specificity of the input. Chain with Competency Unpacker if the outcome needs clarifying before unit design.

3. The performance task is designed for a specific context and may need adaptation. The "Island" scenario works for this Biology topic but may not transfer directly to other schools' resources or assessment requirements. Teachers should review the performance task against their specific assessment framework and modify the scenario while maintaining the assessment design principles.

Competency Unpacker

What This Skill Does

Takes a standard, learning objective, or competency descriptor — often written in abstract, compressed language — and unpacks it into four actionable components: observable indicators (what a student who has achieved this actually DOES), prerequisite knowledge (what must be in place first), common misconceptions (what typically goes wrong), and success criteria at multiple levels (beginning through extending). The output transforms opaque curriculum language into concrete, assessable, teachable components. AI is specifically valuable here because competency descriptors are deliberately compressed — a single sentence like "analyse how writers use language and structure to achieve effects" contains multiple skills, knowledge domains, and levels of sophistication that must be unpacked before they can be taught or assessed.

Evidence Foundation

Wiggins & McTighe (1998, 2005) established that effective curriculum design begins with clarity about desired results — and that most curriculum standards require significant "unpacking" before they can be translated into instruction and assessment. A standard that says "students will understand the causes of World War I" is not assessable until "understand" is defined in observable terms. Marzano & Kendall (2007) provided a taxonomy for classifying the cognitive demands of standards — distinguishing retrieval, comprehension, analysis, and knowledge utilisation — enabling teachers to identify what type of thinking a standard actually requires. Heritage (2008) and Popham (2007) demonstrated that unpacking standards into learning progressions — sequences of sub-skills from prerequisite to target — is essential for both instruction and formative assessment, because it reveals where students are and what they need next. Hattie (2009) found that clear success criteria (effect size 0.77) are among the highest-leverage instructional strategies, but only when they describe what success looks like in specific, observable terms — not when they restate the learning objective in different words.

Input Schema

The teacher must provide:

• Competency descriptor: The standard or objective to unpack. e.g. "Analyse how writers use language and structure to achieve effects and influence readers" / "Use ratio and proportion, including rates, to solve problems" / "Explain how changes in the environment may affect organisms"
• Student level: Year group. e.g. "Year 9"

Optional (injected by context engine if available):

• Subject area: The curriculum subject
• Curriculum framework: Specific standards framework (e.g., National Curriculum, IB, ACARA)
• Student profiles: Prior attainment data, common gaps
• Assessment purpose: Why the competency is being unpacked

Prompt

You are an expert in curriculum design and standards-based instruction, with deep knowledge of Wiggins & McTighe's (1998, 2005) Understanding by Design framework, Marzano & Kendall's (2007) taxonomy of educational objectives, and Heritage's (2008) work on learning progressions. You understand that competency descriptors are compressed summaries of complex learning — they must be unpacked into observable, teachable, assessable components before they can guide instruction.

Your task is to unpack:

**Competency descriptor:** {{competency_descriptor}}
**Student level:** {{student_level}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Subject area:** {{subject_area}} — if not provided, infer from the competency descriptor.
**Curriculum framework:** {{curriculum_framework}} — if not provided, unpack in general terms applicable across frameworks.
**Student profiles:** {{student_profiles}} — if not provided, assume a typical mixed-ability class at the stated year level.
**Assessment purpose:** {{assessment_purpose}} — if not provided, unpack for general instructional planning.

Apply these evidence-based principles:

1. **Observable indicators (Wiggins & McTighe, 2005; Hattie, 2009):**
   - Define what a student who has achieved the competency actually DOES — observable actions, not internal states.
   - "Understands" is not observable. "Identifies," "explains," "compares," "evaluates," "constructs" are observable.
   - Be specific to the competency: "identifies the effect of a writer's word choice on the reader" is more useful than "analyses language."
   - Include 4–6 specific indicators that collectively cover the competency.

2. **Prerequisite knowledge and skills (Heritage, 2008; Popham, 2007):**
   - What must a student already know or be able to do BEFORE they can work toward this competency?
   - Distinguish between content prerequisites (knowledge needed) and skill prerequisites (abilities needed).
   - These prerequisites become diagnostic checks — if a student can't demonstrate the competency, the prerequisite list helps identify WHERE the gap is.

3. **Common misconceptions (Marzano & Kendall, 2007):**
   - What typically goes wrong when students attempt this competency?
   - Include both conceptual misconceptions (misunderstandings of the content) and procedural errors (mistakes in the process).
   - These become assessment design priorities — good assessments surface these misconceptions.

4. **Multi-level success criteria (Hattie, 2009):**
   - Provide success criteria at four levels: Emerging, Developing, Competent, Extending.
   - Each level should describe what the student CAN do (not what they can't).
   - Levels should be qualitatively different, not just quantitatively different (Competent is not "more of" Developing — it's a different quality of response).
   - Use descriptive language, not evaluative labels.

Return your output in this exact format:

## Competency Unpacked: [Brief descriptor]

**Standard:** [Full competency descriptor]
**For:** [Student level]
**Subject:** [Subject area]

### Observable Indicators

[4–6 specific, observable behaviours that demonstrate this competency]

### Prerequisite Knowledge and Skills

**Content prerequisites (what students must know):**
[Specific knowledge required]

**Skill prerequisites (what students must be able to do):**
[Specific skills required]

### Common Misconceptions

[3–4 typical misconceptions or errors, with explanation of why they occur]

### Success Criteria

| Level | What the student does |
|-------|----------------------|
| **Emerging** | [Description of early, partial demonstration] |
| **Developing** | [Description of growing but inconsistent demonstration] |
| **Competent** | [Description of consistent, accurate demonstration] |
| **Extending** | [Description of sophisticated, independent application] |

### Implications for Teaching

[2–3 key teaching implications — what the unpacking reveals about how this competency should be taught]

**Self-check before returning output:** Verify that (a) all indicators are observable, not internal ("explains" not "understands"), (b) prerequisites are specific enough to be diagnostic, (c) misconceptions are based on common patterns, not hypothetical, (d) success criteria describe what students CAN do at each level, and (e) the four levels are qualitatively distinct.

Example Output

Scenario: Competency descriptor: "Analyse how writers use language and structure to achieve effects and influence readers" / Student level: "Year 9" / Subject area: "English Language"

---

Competency Unpacked: Analysing Writers' Language and Structure

Standard: Analyse how writers use language and structure to achieve effects and influence readers For: Year 9 Subject: English Language

Observable Indicators

A student who has achieved this competency:

1. Identifies specific language choices (individual words, phrases, figurative language, sentence types) and names the technique where applicable (metaphor, sibilance, short sentence for impact).
2. Explains the effect of each language choice on the reader — what it makes the reader think, feel, or picture — rather than simply identifying the technique.
3. Identifies structural choices (paragraph length, opening, ending, shift in tone, chronological vs. non-linear, sentence positioning) and explains how they shape the reader's experience across the whole text.
4. Links language and structural choices to the writer's purpose — explains WHY the writer made these choices, connecting technique to intention (to persuade, to create tension, to evoke sympathy, to shock).
5. Uses appropriate analytical terminology accurately — not just naming techniques but using the vocabulary of analysis: "suggests," "conveys," "implies," "positions the reader to."
6. Supports analysis with embedded quotations selected for their analytical potential, not just their availability.

Prerequisite Knowledge and Skills

Content prerequisites (what students must know):

• A working vocabulary of literary and linguistic techniques (metaphor, simile, personification, alliteration, repetition, rhetorical question, short sentence, listing) — at minimum 10 techniques they can identify on sight
• Understanding that writers make deliberate choices — that language is crafted, not accidental
• Awareness that different genres have different purposes (persuade, inform, describe, argue, narrate) and that technique serves purpose

Skill prerequisites (what students must be able to do):

• Read a text extract and identify individual words or phrases that seem important or interesting (close reading at word level)
• Write in paragraphs with a clear point supported by evidence
• Distinguish between describing what happens in a text (summary) and explaining how the text is written (analysis)

Common Misconceptions

1. "Feature-spotting without analysis." The most common error. Students identify a technique ("There is a metaphor") but don't explain its effect. They have been taught to NAME techniques but not to ANALYSE them. This produces responses like: "The writer uses alliteration in 'dark, dangerous, deathly.' This is effective." The identification is correct but the analysis is absent.

2. "The writer uses this to make the reader want to read on." A generic effect claim applied to every technique. Students use this phrase as a default when they don't know what the actual effect is. It signals that the student can identify that techniques have effects but cannot specify them.

3. "Language analysis without structural analysis." Students focus on individual words and phrases (language) but ignore how the text is organised (structure). They analyse the metaphor in paragraph 3 but don't notice that paragraph 3 is a single-sentence paragraph positioned after a long descriptive passage — a structural choice that creates impact.

4. "Technique as the analysis." Students write: "The writer uses a metaphor to show that..." as if naming the technique IS the analysis. The technique is the WHAT; the analysis is the HOW and WHY. Naming "metaphor" is identification; explaining how the metaphor creates a specific image or emotion is analysis.

Success Criteria

Level	What the student does
Emerging	Identifies some language features (e.g., "The writer uses a simile"). May attempt to describe effect but defaults to generic claims ("This is effective" / "This makes the reader want to read on"). Focuses on language only; structural analysis is absent or superficial. Quotations are included but not embedded in analysis.
Developing	Identifies language features AND begins to explain specific effects ("The simile 'cold as a gravestone' suggests death and creates an unsettling atmosphere"). Attempts structural analysis but may describe rather than analyse ("The text starts with a question"). Evidence is relevant but analysis is inconsistent — some points are developed, others stop at identification.
Competent	Analyses both language AND structure with specific, developed explanations of effect. Links techniques to the writer's purpose ("Priestley uses the short sentence 'We are responsible' at the end of the Inspector's speech to deliver his moral message with force — the simplicity contrasts with Birling's verbose speeches, positioning the reader to trust the Inspector's clarity over Birling's bluster"). Quotations are embedded and well-selected. Analysis is consistent across the response.
Extending	Produces perceptive, original analysis that considers multiple possible effects and evaluates which is most likely. Analyses how language and structure work TOGETHER ("The shift from long, flowing sentences to a single blunt statement at the paragraph's end mirrors the character's emotional shift from denial to realisation — the structure enacts what the language describes"). May consider alternative interpretations. Analytical vocabulary is precise and varied.

Implications for Teaching

1. The biggest gap is between identification and analysis. Most students can learn to spot techniques; far fewer can explain effects. Instruction should spend proportionally more time on "What effect does this create?" than on "What technique is this?" Model the difference explicitly: show a feature-spotted response and an analytical response side by side.

2. Structural analysis must be taught separately from language analysis. Students default to word/phrase-level analysis because it's concrete and visible. Structure (paragraph organisation, text-level patterns, positioning of key moments) is more abstract and requires a different analytical lens — looking at the text as a whole, not zooming into individual phrases.

3. Success criteria should be shared with students BEFORE they write. The four-level descriptions can be used as self-assessment tools: "Read your response. Are you at Developing (naming techniques but not always explaining effects) or Competent (consistently explaining HOW techniques create specific effects)? What would you need to add to move up?"

---

Known Limitations

1. The unpacking reflects general expectations for this competency at the stated level. Specific exam boards, curricula, or school policies may define the competency differently or emphasise different components. Teachers should cross-reference the unpacking with their specific assessment framework and adjust where needed.

2. Success criteria at four levels necessarily simplify a continuum. Student work exists on a spectrum, not in neat categories. Some responses may demonstrate Competent analysis of language but Emerging analysis of structure. The levels are guides for feedback, not rigid classifications — the teacher must use professional judgement when a response spans multiple levels.

3. The common misconceptions listed are the MOST common, not all possible ones. Individual students may hold different misconceptions based on their prior instruction, first language, or conceptual framework. The listed misconceptions should be treated as starting points for diagnostic assessment, not as an exhaustive list.

Criterion-Referenced Rubric Generator

What This Skill Does

Produces a criterion-referenced rubric from a learning objective and task description, with descriptive (not evaluative) language at each performance level. Each criterion describes what the student's work LOOKS LIKE at each level — not how "good" it is. The output includes the full rubric, a design rationale, a student-friendly version for self/peer assessment, and calibration notes for consistency across markers. AI is specifically valuable here because effective rubric design requires precise, descriptive language that distinguishes between performance levels without using evaluative labels ("excellent," "good," "poor") or vague quantity indicators ("some," "many," "thorough") — and each descriptor must be qualitatively distinct from the adjacent levels, not just a scaled version of the same description.

Evidence Foundation

Brookhart (2013) established that effective rubrics use descriptive rather than evaluative language — they describe what is PRESENT in the work, not how good it is. "Uses specific textual evidence to support each analytical point" is descriptive; "Good use of evidence" is evaluative. Descriptive rubrics produce more reliable scoring and more useful feedback because they tell students exactly what to do differently, not just that they need to "do better." Andrade (2000, 2013) demonstrated that rubrics improve both instruction and learning when shared with students before the task — they function as learning tools, not just grading tools. The effect is strongest when rubrics are used for self-assessment. Jonsson & Svingby (2007) found that analytic rubrics (separate criteria scored independently) are more reliable and produce better feedback than holistic rubrics (single overall judgment), though they take longer to use. Sadler (1989) established that assessment quality depends on the "gap" being visible — students must be able to see the difference between where they are and where they need to be. Descriptive rubric levels make this gap concrete. Panadero & Jonsson (2013) confirmed that rubric use improves student performance, particularly when combined with self-assessment, with moderate effect sizes.

Input Schema

The teacher must provide:

• Learning objective: What the rubric assesses. e.g. "Students can write a persuasive speech that uses rhetorical devices to influence the audience" / "Students can design and carry out a fair test and draw valid conclusions"
• Task description: The specific task. e.g. "Write and deliver a 3-minute persuasive speech on a topic of your choice" / "Plan and carry out an experiment investigating the effect of light on plant growth, then write a conclusion"
• Student level: Year group. e.g. "Year 8"

Optional (injected by context engine if available):

• Criteria count: Number of criteria (default: 4)
• Performance levels: Number of levels (default: 4)
• Subject area: The curriculum subject
• Existing criteria: Criteria the teacher wants included

Prompt

You are an expert in assessment design and rubric construction, with deep knowledge of Brookhart's (2013) principles of effective rubric design, Andrade's (2000, 2013) research on rubrics as learning tools, and Sadler's (1989) formative assessment framework. You understand that effective rubrics use DESCRIPTIVE language (describing what is present) rather than EVALUATIVE language (judging how good it is), and that each performance level must be QUALITATIVELY distinct — not just a scaled version of the same description.

Your task is to generate a rubric for:

**Learning objective:** {{learning_objective}}
**Task description:** {{task_description}}
**Student level:** {{student_level}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Criteria count:** {{criteria_count}} — if not provided, use 4 criteria.
**Performance levels:** {{performance_levels}} — if not provided, use 4 levels.
**Subject area:** {{subject_area}} — if not provided, infer from the learning objective.
**Existing criteria:** {{existing_criteria}} — if provided, incorporate these and add others as needed.

Apply these evidence-based principles:

1. **Descriptive, not evaluative language (Brookhart, 2013):**
   - NEVER use: excellent, good, satisfactory, poor, limited, basic, outstanding, weak.
   - NEVER use vague quantifiers: some, many, few, several, thorough, adequate.
   - INSTEAD: describe specifically what is PRESENT in the work at each level.
   - Bad: "Good use of evidence." Good: "Each analytical point is supported by a specific, relevant quotation from the text."
   - Bad: "Limited vocabulary." Good: "Vocabulary is predominantly everyday; academic or technical terms are absent or used inaccurately."

2. **Qualitatively distinct levels (Sadler, 1989; Brookhart, 2013):**
   - Adjacent levels must describe qualitatively DIFFERENT work, not just more or less of the same thing.
   - Bad progression: "Uses some evidence" → "Uses more evidence" → "Uses a lot of evidence" (quantitative — just more).
   - Good progression: "States claims without evidence" → "Includes evidence but does not connect it to claims" → "Connects evidence to claims with explicit reasoning" → "Selects the most compelling evidence and explains why it is the strongest support for each claim" (qualitative — different type of work).

3. **Criteria must be independent (Jonsson & Svingby, 2007):**
   - Each criterion should assess a distinct, separable aspect of the work.
   - A student should be able to score at different levels on different criteria — not all criteria tracking together.
   - If two criteria always produce the same score, they're probably measuring the same thing.

4. **Student-friendly version (Andrade, 2000, 2013):**
   - Produce a version using student-accessible language.
   - Written in second person: "You support each point with a specific quotation" rather than "The student supports each point with a specific quotation."
   - Can be used for self-assessment before submission.

5. **Calibration notes (Jonsson & Svingby, 2007):**
   - Include notes that help multiple markers apply the rubric consistently.
   - Identify the key distinctions between levels — the "decision points" where markers are most likely to disagree.
   - Where possible, describe what a borderline response looks like between two levels.

Return your output in this exact format:

## Rubric: [Task Description]

**Assesses:** [Learning objective]
**For:** [Student level]
**Criteria:** [Count] | **Levels:** [Count]

### Rubric

[Full rubric table with criteria as rows and performance levels as columns. Each cell contains descriptive language.]

### Design Rationale

[Why these criteria were chosen and how the levels are distinguished]

### Student-Friendly Version

[Simplified rubric in student-accessible language, suitable for self/peer assessment]

### Calibration Notes

[Guidance for consistent application — key decision points, borderline cases]

**Self-check before returning output:** Verify that (a) NO evaluative language appears (no "good," "excellent," "poor," "limited"), (b) NO vague quantifiers appear (no "some," "many," "thorough"), (c) adjacent levels are qualitatively distinct, (d) each criterion is independent, (e) the student-friendly version is genuinely accessible, and (f) calibration notes address the most likely points of marker disagreement.

Example Output

Scenario: Learning objective: "Students can write a persuasive speech that uses rhetorical devices to influence the audience" / Task description: "Write and deliver a 3-minute persuasive speech on a topic of your choice to your class" / Student level: "Year 9" / Subject area: "English"

---

Rubric: Persuasive Speech (Written and Delivered)

Assesses: Write a persuasive speech that uses rhetorical devices to influence the audience For: Year 9 English Criteria: 4 | Levels: 4

Rubric

Criterion	Level 1	Level 2	Level 3	Level 4
Argument and Structure	The speech states an opinion but does not develop it with reasons. Points are listed without logical connection. There is no clear beginning, middle, or end — the speech could start or stop at any point without the audience noticing.	The speech states a position and provides 1–2 reasons, but the reasons are not fully developed. The speech has a recognisable opening and closing, but the middle section lacks clear organisation — points do not build on each other.	The speech presents a clear position supported by 3+ developed reasons arranged in a logical sequence. Each point builds on the previous one, creating a cumulative argument. The opening engages the audience and the conclusion reinforces the position without simply repeating it.	The speech constructs a strategic argument where the ordering of points is deliberate — the strongest point is positioned for maximum impact. Counter-arguments are acknowledged and addressed. The conclusion reframes the argument in a way that leaves the audience with a lasting impression.
Rhetorical Devices	The speech relies on personal opinion ("I think...") without any deliberate use of rhetorical techniques. Language is conversational throughout.	The speech includes 1–2 rhetorical devices (e.g., rhetorical question, repetition, rule of three) but they appear isolated — inserted into conversational language rather than integrated into the argument. The devices are present but do not noticeably strengthen the persuasion.	The speech uses 3+ rhetorical devices that are integrated into the argument — each device serves a specific persuasive purpose (a rhetorical question to challenge the audience, repetition to emphasise a key point, emotive language to create urgency). The speaker can explain why they chose each device.	The speech demonstrates controlled and varied use of rhetoric — devices are combined for effect (e.g., a rhetorical question followed by a tricolon that answers it). The tone shifts strategically (from rational to emotional, from calm to urgent) to position the audience. Language choices reflect awareness of the specific audience.
Audience Awareness	The speech is written for a general audience — there are no specific references to the listeners, their experiences, or their likely objections. The speaker talks AT the audience rather than TO them.	The speech includes some awareness of the audience (e.g., "As students, we all know...") but this awareness is surface-level — it acknowledges the audience without adapting the argument to their specific concerns or perspectives.	The speech directly addresses the audience's likely views, concerns, or experiences. The speaker anticipates what the audience might think or feel and addresses it: "You might be thinking... but consider this." The speech feels directed at THESE listeners, not any audience.	The speech is strategically tailored to the specific audience — it uses shared references, anticipates specific objections, and adjusts register, vocabulary, and tone to match the listeners. The speaker reads the room during delivery and responds to audience reactions (adjusting pace, adding emphasis, making eye contact with sceptics).
Delivery	The speech is read from a script with minimal eye contact. Volume is either too quiet to hear or monotone throughout. The speaker appears disconnected from their own argument.	The speech is mostly read with occasional glances at the audience. Volume is audible. Some variation in tone is attempted, but pacing is uneven — the speaker rushes through some sections and pauses awkwardly in others.	The speaker uses notes as prompts rather than reading a full script. Eye contact is regular and distributed across the audience. Volume, pace, and emphasis vary deliberately to reinforce key points. Pauses are used for effect, not from uncertainty.	The speaker is largely independent of notes. Delivery feels natural and confident — the speaker appears to be thinking and speaking, not performing a memorised text. Gestures, movement, facial expression, and vocal variety all serve the argument. The speaker commands the room.

Design Rationale

Why these four criteria:

1. Argument and Structure — the foundation. A speech without a clear argument is not persuasive, regardless of how well it's delivered.
2. Rhetorical Devices — the tool kit. The learning objective specifically requires rhetorical devices; this criterion assesses whether they're present AND functional.
3. Audience Awareness — the persuasive lens. Persuasion is fundamentally about the audience; this criterion assesses whether the speaker writes FOR their listeners or just AT them.
4. Delivery — the performance dimension. A speech is a spoken genre; content quality and delivery quality are separate skills that should be assessed independently. A student might write a brilliant speech and deliver it poorly, or vice versa — separate criteria capture both.

How levels are distinguished:

• Level 1 → Level 2: The element is PRESENT but not effective (devices exist but don't persuade; structure exists but doesn't build).
• Level 2 → Level 3: The element is FUNCTIONAL — it serves a purpose and achieves an effect (devices are integrated, structure builds, audience is directly addressed).
• Level 3 → Level 4: The element is STRATEGIC — the student makes deliberate, sophisticated choices and can explain them (rhetoric is combined, argument is ordered for impact, audience is read and responded to).

Student-Friendly Version

Before you submit, check your speech against these criteria:

Your argument:

• Level 1: You state your opinion but don't really explain why.
• Level 2: You give reasons but they're not fully developed — you haven't explained WHY your reasons are convincing.
• Level 3: You present a clear argument with 3+ reasons that build on each other. Your opening grabs attention and your ending reinforces your point.
• Level 4: You've arranged your points strategically — putting your strongest point where it has the most impact. You address what the other side would say.

Your use of rhetorical devices:

• Level 1: You haven't used any persuasive techniques — it just sounds like you're talking normally.
• Level 2: You've used 1–2 techniques but they feel "stuck on" — like you added them because you had to.
• Level 3: You've used 3+ techniques and each one has a PURPOSE — you can explain why you chose it and what effect it creates.
• Level 4: Your techniques work TOGETHER — you combine them, and you change your tone deliberately to affect the audience.

Your awareness of the audience:

• Level 1: You're talking AT your audience, not TO them.
• Level 2: You mention the audience ("As students, we...") but don't really adapt your argument to them.
• Level 3: You directly address what your audience thinks and feels. "You might be thinking... but consider this."
• Level 4: Your whole speech is shaped for THIS specific audience — you use their language, anticipate their objections, and adjust as you go.

Your delivery:

• Level 1: You're reading your speech with your eyes on the paper.
• Level 2: You look up sometimes. Your voice is audible but doesn't really change.
• Level 3: You use notes, not a full script. You make eye contact. You deliberately change speed, volume, and emphasis.
• Level 4: You barely need notes. Your delivery feels natural and confident. Your body language and voice both serve your argument.

Calibration Notes

Key decision points for markers:

1. Level 2 vs. Level 3 on Rhetorical Devices: The distinction is between devices that are PRESENT and devices that are FUNCTIONAL. Ask: "If I removed this rhetorical question, would the speech be less persuasive?" If yes → Level 3 (it's doing work). If no → Level 2 (it's decorative).

2. Level 3 vs. Level 4 on Argument and Structure: Level 3 has a clear, logical argument. Level 4 has a STRATEGIC argument — the ordering is deliberate, and counter-arguments are addressed. If the speech presents good points but in no particular order → Level 3. If the ordering itself reveals strategic thinking (building to a climax, saving the strongest point) → Level 4.

3. Delivery is independent of content. A student may deliver a poorly structured speech with great confidence (high Delivery, low Argument) or a brilliant speech in a monotone (high Argument, low Delivery). Score each criterion independently.

4. Borderline between Level 2 and Level 3: The most common borderline. If in doubt, ask: "Is the student just doing the thing (using a device, addressing the audience) or is the student doing the thing WITH PURPOSE (using the device to achieve a specific effect, addressing the audience's specific concerns)?" Purpose → Level 3. Going through the motions → Level 2.

---

Known Limitations

1. Rubrics describe performance but do not explain how to improve. A student who reads the rubric and sees they are at Level 2 for Rhetorical Devices knows WHAT to do differently (integrate devices with purpose) but may not know HOW. The rubric should be paired with teaching and feedback that shows students how to move from one level to the next. Chain with Feedback Quality Analyser for targeted improvement advice.

2. Four levels is a practical compromise. Some tasks would benefit from more levels (to distinguish fine gradations) or fewer (to simplify assessment). Four levels balance reliability (enough levels to be informative) with usability (few enough to be practical). If the rubric is being used for high-stakes grading, additional level descriptors may be needed.

3. Descriptive language is harder to write but more useful than evaluative language. The rubric avoids "good," "excellent," and "poor," which makes each cell longer and more specific. This is a deliberate trade-off — evaluative rubrics are shorter but less useful for feedback. Teachers may need time to become comfortable with descriptive rubric language.

Curriculum Knowledge Architecture Designer

What This Skill Does

Takes a curriculum input — a single course, a subject scope and sequence, or a real-world project brief — and diagnoses the epistemic architecture of the knowledge domain. It determines whether the domain is primarily Hierarchical, Horizontal, Dispositional, or a mixed architecture, then constructs the appropriate knowledge structure map for each type present, and outputs concrete implications for teaching sequence, assessment design, and AI tutoring architecture. Most real curricula — especially project-based and real-world learning designs — are mixed architectures. The skill diagnoses proportion and interaction, not forcing a single type. AI is specifically valuable here because epistemic diagnosis requires simultaneously applying sociological theory (Bernstein's knowledge structures), curriculum design expertise (sequencing and assessment logic), and competency framework literacy (dispositional progression) — a combination that is rare in any single educator and time-consuming to work through manually.

Evidence Foundation

Bernstein (1999) distinguished two forms of discourse — horizontal discourse (everyday, context-specific knowledge) and vertical discourse (systematic, principled knowledge) — and within vertical discourse identified two knowledge structures. Hierarchical knowledge structures are coherent, explicitly principled, and hierarchically integrated: new theory subsumes and generalises prior knowledge, creating a cumulative progression where lower-level concepts must be mastered before higher-level ones are accessible. The natural sciences are the paradigmatic example. Horizontal knowledge structures are organised as a series of specialised languages or lenses, each with its own modes of inquiry and criteria for valid knowledge. Development occurs through accumulation of new perspectives rather than integration. The humanities and social sciences are paradigmatic. Bernstein (2000) extended this framework through the concept of recontextualisation — how knowledge is transformed as it moves from its field of production into pedagogic contexts — which directly informs how curriculum designers must think about knowledge type when making sequencing decisions.

Muller (2009) applied Bernstein's framework to curriculum coherence, distinguishing conceptual coherence (characteristic of hierarchical knowledge — curricula where knowledge builds cumulatively on prior knowledge) from contextual coherence (characteristic of segmental curricula — where each segment is adequate to a specific context but segments do not necessarily build on one another). This distinction has direct implications for sequencing: conceptually coherent curricula have a logic that is difficult to reorder, while contextually coherent curricula can be entered from multiple points.

Maton (2009, 2013, 2014) developed Legitimation Code Theory's Semantics dimension, providing two analytical tools: semantic gravity (the degree to which meaning is tied to a specific context — stronger SG means more contextual, weaker SG means more abstract and transferable) and semantic density (the degree to which meaning is condensed into terms or symbols). Maton (2013) introduced the concept of semantic waves — the pedagogic practice of moving between concrete examples (high SG, low SD) and abstract principles (low SG, high SD) — showing that curricula and teaching that create these waves enable cumulative knowledge-building, while those that remain flat (always contextual or always abstract) produce segmented learning. This provides a diagnostic tool for identifying where in a curriculum conceptual unpacking and repacking are needed.

Young (2008) argued that curriculum theory must take seriously which knowledge matters — introducing the concept of powerful knowledge: specialised, systematic, discipline-based knowledge that gives learners access to explanatory frameworks they cannot acquire through everyday experience. Wheelahan (2010) extended this argument to show that competency-based curricula that strip knowledge down to contextual skills without theoretical grounding deny students access to the conceptual structures that enable social participation — making knowledge architecture a question of equity, not merely pedagogy.

The dispositional knowledge category draws on competency framework literature. Unlike hierarchical and horizontal structures (which describe how propositional knowledge is organised), dispositional knowledge is constituted by developing capacities, orientations, and enacted competencies — it exists only in enactment. The EU competency frameworks provide the most rigorous articulations: GreenComp (Bianchi, Pisiotis & Cabrera Giraldez, 2022) defines twelve sustainability competences including agency, systems thinking, and values literacy; EntreComp (Bacigalupo et al., 2016) defines fifteen entrepreneurship competences including self-awareness, creativity, and learning through experience across an eight-level progression model; LifeComp (Sala et al., 2020) defines nine personal, social, and learning-to-learn competences including self-regulation, collaboration, and critical thinking. These frameworks share a common characteristic: progression is qualitative and developmental, described through bands rather than prerequisite chains, and assessment requires teacher judgment of enacted capability rather than testing of propositional knowledge.

Input Schema

The teacher must provide:

• Curriculum input type: What kind of input are you providing? e.g. "course" / "scope-and-sequence" / "project-brief"
• Domain or subject: Name and brief description. e.g. "Year 9 Science — forces, energy, and motion" / "Design and Regeneration project — smart water systems for a local wetland" / "IB History — causes and consequences of 20th-century authoritarian states"
• Learner stage: Age range or year group. e.g. "12–14 years" / "Year 10" / "Grade 7–8"
• Learning goals: What students should know, understand, and be able to do — 3–5 sentences. e.g. "Students will understand how forces cause changes in motion, apply Newton's laws to real-world situations, design and test a simple machine, and communicate scientific reasoning through written argument."

Optional (injected by context engine if available):

• Existing curriculum documents: Text of curriculum documents, unit plans, or scope-and-sequence
• Competency framework: The school's dispositional or competency framework
• Prior knowledge baseline: What students already know and can do

Prompt

You are an expert in curriculum epistemology and knowledge structure analysis, with deep knowledge of Bernstein's (1999, 2000) theory of knowledge structures, Muller's (2009) work on curriculum coherence, and Maton's (2009, 2013, 2014) Legitimation Code Theory — particularly the Semantics dimension (semantic gravity and semantic density). You also understand competency framework design (GreenComp, EntreComp, LifeComp) and the distinction between propositional knowledge structures and dispositional development.

Your task is to diagnose the epistemic architecture of the following curriculum input and produce a complete knowledge architecture analysis.

**Curriculum input type:** {{curriculum_input_type}}
**Domain or subject:** {{domain_or_subject}}
**Learner stage:** {{learner_stage}}
**Learning goals:** {{learning_goals}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Existing curriculum documents:** {{existing_curriculum_documents}} — if not provided, work from the domain, subject description, and learning goals.
**Competency framework:** {{competency_framework}} — if not provided, identify any dispositional elements from the learning goals.
**Prior knowledge baseline:** {{prior_knowledge_baseline}} — if not provided, assume typical prior knowledge for this learner stage.

Apply the following framework. You MUST use the exact three knowledge types defined below. Most real curricula contain more than one type — your diagnosis must identify ALL types present, estimate their approximate proportion, and explain how they interact.

## The Three Knowledge Types

**1. Hierarchical Knowledge Structure (Bernstein, 1999)**
Knowledge is coherent, systematically principled, and organised so that lower-level concepts must be mastered before higher-level ones are accessible. Development is through integration, where new theory subsumes and generalises prior knowledge. The curriculum logic is cumulative — sequencing is constrained by prerequisite relationships. Conceptual coherence (Muller, 2009) holds: content cannot be freely reordered without loss.

Indicators: prerequisite chains exist; concepts build in abstraction; mastery of prior concepts is necessary for later ones; errors at lower levels propagate upward; the subject has a canonical sequencing logic.

Examples: mathematics, formal logic, music theory, programming fundamentals, chemistry, physics.

**2. Horizontal Knowledge Structure (Bernstein, 1999)**
Knowledge is organised as a series of specialised languages or lenses, each with its own modes of inquiry and criteria for valid knowledge. Content can be entered from multiple points; development is through accumulation of new perspectives rather than integration. Disciplinary thinking skill develops progressively even though content is not strictly prerequisite-ordered. Contextual coherence (Muller, 2009) allows curriculum segments to be reordered, though analytical sophistication still develops cumulatively.

Indicators: multiple valid interpretive lenses exist; content can be studied in various orders; "thinking like a historian/philosopher/sociologist" develops across the curriculum rather than through a fixed sequence; new units add perspectives rather than building on prior units.

Examples: history, literature, philosophy, geography, sociology, art criticism.

**3. Dispositional Knowledge Structure**
Knowledge is constituted by developing capacities, orientations, and enacted competencies. The knowledge cannot be separated from the learner's growing capability — it exists only in enactment. Progression is qualitative and developmental, requiring teacher judgment rather than automated assessment. Competency frameworks (GreenComp, EntreComp, LifeComp) provide the most rigorous articulations of this type.

Indicators: the learning goal describes who the student is becoming, not just what they know; progression is described in developmental bands (emerging → developing → extending) rather than prerequisite chains; assessment requires observation of enacted behaviour over time; the competency cannot be tested through a single task.

Examples: agency, collaboration, ecological literacy, entrepreneurial thinking, self-regulation, creative confidence, regenerative mindset.

## Diagnosis Process

Step 1: Read the curriculum input carefully. For each learning goal, determine which knowledge type(s) it belongs to.

Step 2: Estimate the approximate proportion of each type present (as percentages that sum to 100%). Explain your reasoning — which specific goals or content areas belong to which type.

Step 3: For each type present, build the appropriate knowledge structure map:
- **Hierarchical:** Identify the key prerequisite chains. Order concepts topologically — which concepts must come before which. Flag any concepts where the prerequisite relationship is hard (cannot proceed without it) vs soft (easier with it but possible without).
- **Horizontal:** Identify the conceptual hubs (central themes, phenomena, or questions) and the lenses or perspectives that orbit each hub. Show how analytical sophistication develops across the curriculum even though content is not prerequisite-ordered.
- **Dispositional:** Define progression band descriptors across 4 levels: Emerging → Developing → Competent → Extending. Each level must describe what the learner DOES (observable behaviour), not what they "understand" internally. Include indicators that distinguish between levels.

Step 4: Analyse the mixed architecture — where do types interact, overlap, or create tension? Where does a hierarchical prerequisite chain intersect with a dispositional development goal? Where does a horizontal lens require hierarchical foundational knowledge?

Step 5: Derive implications for teaching, assessment, and AI tutoring.

Return your output in this exact format:

## Knowledge Architecture Analysis: [Domain/Subject Name]

**Input type:** [course / scope-and-sequence / project-brief]
**Learner stage:** [age/year]
**Learning goals:** [Summarised]

### 1. Epistemic Diagnosis

**Architecture type:** [Mixed / Primarily Hierarchical / Primarily Horizontal / Primarily Dispositional]
**Proportions:** [e.g. 40% Hierarchical, 35% Horizontal, 25% Dispositional]

**Reasoning:**
[For each knowledge type present, explain which specific learning goals or content areas belong to it and why. Reference the indicators from the framework above.]

### 2. Knowledge Architecture Map

#### Hierarchical Elements
[If present: prerequisite chains with topological ordering. Show which concepts must come before which. Mark hard vs soft prerequisites. Use a visual chain format.]

#### Horizontal Elements
[If present: conceptual hubs and the lenses/perspectives that orbit them. Show how analytical thinking develops across the curriculum.]

#### Dispositional Elements
[If present: progression band descriptors across 4 levels — Emerging, Developing, Competent, Extending — with observable indicators at each level for each dispositional competency identified.]

### 3. Mixed Architecture Notes

[Where do types interact? Where does tension arise? Be specific to THIS curriculum.]

### 4. Teaching & Sequencing Implications

[What does the architecture mean for how content should be ordered and paced? Which elements have constrained sequencing (hierarchical) vs flexible sequencing (horizontal)? Where must dispositional development run as a continuous thread rather than being assigned to specific lessons?]

### 5. Assessment Implications

**Auto-assessable elements (suitable for AI/automated assessment):**
[List specific elements and explain why they are auto-assessable — typically hierarchical elements with clear right/wrong answers or demonstrable procedures.]

**Teacher-judgment elements (require human assessment):**
[List specific elements and explain why they require teacher judgment — typically dispositional elements and sophisticated horizontal analysis.]

**Mixed elements (partial automation possible):**
[Elements where AI can assess some dimensions but teacher judgment is needed for others.]

### 6. AI Tutoring Design Implications

[How should this architecture inform an intelligent tutoring system or AI teacher assistant? Consider:
- For hierarchical elements: adaptive sequencing, prerequisite checking, targeted practice
- For horizontal elements: perspective-prompting, analytical scaffolding, exposure to multiple lenses
- For dispositional elements: reflection prompts, portfolio tracking, teacher dashboards (NOT automated grading)
- For mixed architectures: how the tutoring system should handle the transitions between types]

**Self-check before returning output:** Verify that (a) all three knowledge types have been considered, not just the dominant one, (b) proportions are justified with specific references to learning goals, (c) the architecture map uses the correct format for each type (prerequisite chains for hierarchical, hubs-and-lenses for horizontal, progression bands for dispositional), (d) mixed architecture notes are specific to THIS curriculum rather than generic, (e) assessment implications distinguish clearly between auto-assessable and teacher-judgment elements, and (f) AI tutoring implications are practical and architecture-specific.

Example Output

Read references/worked-examples.md when you need the complete Smart Water Systems example, including prerequisite diagrams, conceptual lenses, dispositional progressions and tutoring implications.

Compact excerpt — interpreting a mixed architecture (Smart Water Systems):

Architecture type: Mixed — all three knowledge types are present in significant proportions. Proportions: 40% Hierarchical, 30% Horizontal, 30% Dispositional

Key tension: Assessment of the hierarchical elements (water quality science, electronics) can be standardised and criterion-referenced. Assessment of the dispositional elements (agency, collaboration, regenerative mindset) cannot — it requires sustained observation and professional judgment. A project assessment framework that weights both equally must accommodate fundamentally different modes of evidence. There is a risk that the easily-measured hierarchical elements dominate assessment simply because they are easier to assess, marginalising the dispositional goals that are arguably the most important outcomes of a D2R project.

---

Known Limitations

1. The three-type framework is a simplification. Bernstein's original distinction was between hierarchical and horizontal knowledge structures within vertical discourse. The dispositional category is not Bernstein's — it draws on competency framework literature and represents a pragmatic extension for curriculum design purposes. Scholars working within Bernstein's framework may disagree with treating dispositional knowledge as a separate structural type rather than as a feature of horizontal discourse. The framework is offered as a design tool, not as a contribution to sociological theory.

2. Proportions are approximate and contestable. Estimating that a curriculum is "40% hierarchical, 30% horizontal, 30% dispositional" implies a precision that the framework cannot deliver. The proportions are interpretive judgments — a different analyst might assess the same curriculum as 35/35/30 or 45/25/30. They are useful for surfacing the relative weight of different knowledge types and ensuring none is overlooked, not as exact measurements.

3. The diagnostic is based on stated learning goals, not enacted curriculum. A curriculum that states dispositional goals (agency, collaboration) may not actually develop them if the teaching and assessment prioritise only the hierarchical elements. The architecture analysis reveals what the curriculum CLAIMS to do; whether it actually does so depends on implementation. Teachers should use the analysis to check whether their assessment design and daily teaching practice genuinely address all the knowledge types their curriculum claims to include.

4. Knowledge-contingent dispositions require prerequisite diagnosis, not just type labelling. Some dispositional goals — particularly critical thinking, ecological literacy, regenerative mindset, and entrepreneurial thinking — cannot develop authentically without a sufficient hierarchical and horizontal knowledge base. The architecture diagnosis identifies that a dispositional goal is present; it does not automatically identify whether the prerequisites for that disposition to be operable are also present in the curriculum. When the diagnosis identifies a knowledge-contingent disposition, the teaching and assessment implications must include explicit identification of the hierarchical and horizontal elements that must be developed first. A curriculum that states "students will develop critical thinking" without ensuring sufficient domain knowledge to think critically with is not a curriculum that will achieve that goal.

5. Dispositional progression bands describe typical development, not universal stages. A student may demonstrate "Extending" agency in one context and "Emerging" in another. Dispositional development is context-sensitive and non-linear. The bands are guides for observation and feedback, not rigid classifications.

6. The AI tutoring implications assume a system capable of mode-switching. Most current AI tutoring systems are designed for hierarchical knowledge (adaptive practice with right/wrong feedback). The recommendations for horizontal and dispositional elements describe capabilities that most systems do not yet have. The implications are forward-looking — describing what an architecture-informed system SHOULD do, not what off-the-shelf systems currently can do.

Differentiation Adapter

What This Skill Does

Adapts a task for a specific learner profile — extension, support, EAL, ADHD, dyslexia, anxiety, visual impairment, autism, gifted and talented — while explicitly maintaining the same learning objective. The critical principle is that differentiation modifies the ROUTE to learning, not the DESTINATION. A student with dyslexia attempting the same learning objective as their peers may need different input formats, different response formats, and different scaffolding — but they should be working toward the same understanding. The output includes the adapted task, an explicit statement of what changed and what stayed the same, a verification that the learning objective is maintained, and implementation notes. AI is specifically valuable here because effective differentiation requires knowledge of both the learner profile (what barriers does this profile create?) and the task (which elements of this task create those barriers?) — a two-way analysis that must be done for each combination of task and learner need.

Evidence Foundation

Tomlinson (2001, 2014) established the framework for differentiated instruction, identifying three dimensions of differentiation: content (what students learn), process (how they learn it), and product (how they demonstrate learning). She emphasised that differentiation should be by readiness, interest, and learning profile — NOT by learning style (which is excluded from this library as debunked). Rose & Meyer (2002) developed Universal Design for Learning (UDL), arguing that curricula should be designed from the outset to be accessible to all learners through three principles: multiple means of engagement (the "why" of learning), multiple means of representation (the "what"), and multiple means of action and expression (the "how"). Vygotsky (1978) established that instruction should target the Zone of Proximal Development — what the learner can do with appropriate support but not yet independently. Hattie (2009) found that differentiation has moderate effect sizes overall but varies significantly by implementation quality — poorly implemented differentiation (giving weaker students easier work) can actually reduce achievement by lowering expectations. CAST (2018) provided UDL Guidelines version 2.2, the historical edition used here for implementation strategies. CAST released version 3.0 in 2024; this skill retains its stated 2.2 basis rather than claiming to implement the newer edition.

Input Schema

The teacher must provide:

• Original task: The task as designed. e.g. "Read the extract from 'A Christmas Carol' and write a paragraph analysing how Dickens presents Scrooge's transformation, using quotations as evidence."
• Learner profile: The specific need. e.g. "Extension — student who finishes quickly and needs deeper challenge" / "Support — student with dyslexia who struggles with reading-heavy tasks" / "ADHD — student who struggles with sustained focus on extended writing"
• Learning objective: What all students should learn. e.g. "Analyse how Dickens presents character change using textual evidence."

Optional (injected by context engine if available):

• Student level: Year group
• Subject area: The curriculum subject
• Student profiles: Specific diagnoses, support plans, prior attainment
• Available support: TA, technology, specialist resources

Prompt

You are an expert in differentiated instruction and inclusive education, with deep knowledge of Tomlinson's (2001, 2014) differentiation framework, Rose & Meyer's (2002) Universal Design for Learning principles, and CAST's (2018) UDL Guidelines version 2.2. This is the historical edition used here, not a claim to implement version 3.0 (released in 2024). You understand that effective differentiation modifies the ROUTE to learning, not the DESTINATION — all students work toward the same learning objective, but the pathway is adapted to remove barriers specific to each learner's profile.

IMPORTANT: Differentiation by learning style (visual/auditory/kinaesthetic preferences) is NOT supported — this is excluded from this library as debunked (Pashler et al., 2008). Differentiation by readiness, specific learning needs, and learner profile IS supported.

Your task is to adapt:

**Original task:** {{original_task}}
**Learner profile:** {{learner_profile}}
**Learning objective:** {{learning_objective}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Student level:** {{student_level}} — if not provided, infer from the task.
**Subject area:** {{subject_area}} — if not provided, infer from the task.
**Student profiles:** {{student_profiles}} — if not provided, base adaptations on general research about the stated learner profile.
**Available support:** {{available_support}} — if not provided, assume standard classroom resources with no specialist TA.

Apply these evidence-based principles:

1. **Same objective, different route (Tomlinson, 2001):**
   - The learning objective must be identical for the adapted and original task.
   - Modify HOW the student engages with the content or demonstrates their learning, not WHAT they learn.
   - If the adaptation reduces cognitive demand, it has gone too far — it should reduce barriers, not reduce thinking.

2. **Profile-specific adaptations (UDL, CAST 2018):**
   Adapt based on what the research says about each learner profile:
   - **Extension/Gifted:** Increase depth and complexity — not more of the same, but qualitatively different challenge. Abstract thinking, multiple perspectives, evaluation, creation.
   - **Support/Below expected level:** Additional scaffolding — sentence frames, graphic organisers, reduced volume (not reduced difficulty), worked examples to reference.
   - **Dyslexia:** Reduce reading load without reducing thinking. Larger font, coloured overlay, audio version of text, key quotations pre-selected, scribe option for writing. Focus on demonstrating UNDERSTANDING, not reading fluency.
   - **ADHD:** Break task into shorter chunks with check-in points. Reduce unnecessary information. Provide movement breaks. Use timers for focused bursts. Minimise distractions in the task presentation.
   - **EAL:** Language scaffolds (sentence frames, word banks, glossary, bilingual support) — see the EAL domain skills for detailed approaches.
   - **Anxiety:** Reduce performance pressure. Allow draft attempts. Provide clear structure. Offer choice of response format. Avoid cold-calling or public demonstration of work-in-progress.
   - **Autism:** Provide explicit, unambiguous instructions. Avoid figurative language in task instructions (or gloss it). Provide predictable structure. Allow additional processing time.
   - **Visual impairment:** Enlarged text, high contrast, audio alternatives, tactile resources where appropriate.

3. **UDL multiple means (Rose & Meyer, 2002):**
   - **Multiple means of representation:** Can the content be presented differently? (Audio, visual, simplified text alongside original, graphic organiser)
   - **Multiple means of action and expression:** Can the student demonstrate learning differently? (Verbal instead of written, diagram instead of essay, recorded instead of live)
   - **Multiple means of engagement:** Can the task be connected to the student's interests or motivations?

4. **Avoid common differentiation errors (Hattie, 2009):**
   - Do NOT give a simpler version of the task to struggling students — this lowers expectations and reduces learning.
   - Do NOT give "more work" to extension students — depth, not volume.
   - Do NOT assume that the adapted version is inherently "lower" — it should be equally demanding but differently accessible.

Return your output in this exact format:

## Adapted Task: [Learner Profile]

**Original task:** [Brief description]
**Adapted for:** [Learner profile]
**Learning objective:** [Must be identical to original]

### The Adapted Task

[Complete adapted version of the task with all modifications in place]

### What Changed and What Stayed the Same

**Changed:** [Specific modifications — what was adapted and why]
**Maintained:** [What was deliberately kept the same — especially cognitive demand and learning objective]

### Objective Check

[Explicit verification: "The adapted task still requires the student to [learning objective] because..."]

### Implementation Notes

[Practical advice for the teacher: how to introduce the adaptation, how to avoid singling the student out, what to watch for]

**Self-check before returning output:** Verify that (a) the learning objective is identical, (b) adaptations address the specific barriers of the stated learner profile, (c) cognitive demand is maintained — the adapted task is not easier, just more accessible, (d) the adaptation is based on research about the learner profile, not on learning style preferences, and (e) the implementation notes address how to manage the adaptation in a classroom without stigmatising the student.

Example Output

Scenario: Original task: "Read the extract from 'A Christmas Carol' and write a paragraph analysing how Dickens presents Scrooge's transformation, using quotations as evidence." / Learner profile: "Dyslexia — student has a reading age approximately 2 years below chronological age, struggles with decoding longer texts, avoids reading tasks, but demonstrates strong verbal analytical skills in class discussion" / Learning objective: "Analyse how Dickens presents character change using textual evidence" / Student level: "Year 9" / Subject area: "English Literature"

---

Adapted Task: Dyslexia

Original task: Read the extract from 'A Christmas Carol' and write a paragraph analysing how Dickens presents Scrooge's transformation, using quotations as evidence. Adapted for: Dyslexia — strong verbal analysis, reading age ~2 years below chronological age Learning objective: Analyse how Dickens presents character change using textual evidence

The Adapted Task

Reading the extract:

• Provide the extract printed on cream/buff paper (reduces visual contrast stress) in a minimum 14pt font with 1.5 line spacing.
• Pre-highlight 4 key quotations that show Scrooge's transformation — these are the quotations the student should focus on. This removes the barrier of scanning dense text to find evidence (a reading-fluency task) while maintaining the requirement to ANALYSE the evidence (the actual learning objective).
• If audio is available, provide an audio recording of the extract that the student can listen to while following the text. This supports decoding without removing the text entirely.

Analysing and writing:

• Provide a PEEL paragraph frame:
  • Point: "Dickens presents Scrooge's transformation by..."
  • Evidence: "This is shown when Dickens writes '___' (one of the highlighted quotations)"
  • Explain: "This quotation suggests that Scrooge has changed because..."
  • Link: "This shows the reader that..."
• The student must supply their own analytical thinking at each stage — the frame structures the writing, not the analysis.
• Alternative response option: If writing is the primary barrier (not analysis), offer the student the choice to record their paragraph verbally (voice recording or dictation to a scribe) instead of writing it. The analytical demand is identical; only the output format changes.

What Changed and What Stayed the Same

Changed:

• Text presentation: larger font, cream paper, increased spacing — reduces decoding load
• Quotations pre-highlighted: removes the need to scan the full text for evidence — a reading-fluency task, not an analytical task
• Audio support: optional listening alongside reading
• Writing scaffold: PEEL frame provided to structure the paragraph
• Alternative output: verbal response option available

Maintained:

• The learning objective: analyse how Dickens presents character change using textual evidence
• The requirement to use specific quotations as evidence
• The analytical demand: the student must explain WHAT the quotation shows and WHY it's significant
• The quality expectation: the analysis should be just as perceptive as any other student's
• The text itself: the same extract, not a simplified version

Objective Check

The adapted task still requires the student to analyse how Dickens presents character change using textual evidence because:

• The student must still READ the extract (with support, but the same text)
• The student must still SELECT a quotation and explain its significance
• The student must still ANALYSE — explain what the quotation reveals about Scrooge's transformation and why Dickens chose to present it this way
• The pre-highlighting removes a reading-fluency barrier, not an analytical barrier — the student still chooses which highlighted quotation to use and must explain its significance independently
• The PEEL frame structures the writing but does not supply the analytical content — every "Explain" and "Link" section requires the student's own thinking

Implementation Notes

1. Introduce the adaptation discreetly. Don't announce "This is the dyslexia version." Instead, make some adaptations available to all students: print the extract on cream paper for everyone (it helps all readers). Make the audio version available to the whole class. Differentiate the highlighted quotations individually — hand them out rather than projecting.

2. Monitor the analysis, not the writing. This student's strength is verbal analysis. If the written response is weaker than expected, check whether the barrier is writing-mechanics (spelling, handwriting, sentence construction) or analysis. If the student can explain their analysis verbally but not in writing, the barrier is writing production, not understanding — and the voice-recording option should be offered.

3. Don't lower expectations for analytical quality. The student with dyslexia has strong verbal analytical skills. Their analysis should be held to the same standard as any other student's. The adaptation removes a barrier; it does not lower the bar. If the student's written paragraph is thin, the response is "Can you tell me more about what this quotation shows?" — pushing for depth — not "Well done for trying."

4. Revisit the pre-highlighting over time. Pre-selecting quotations is a scaffold that should be reduced as the student develops reading stamina and confidence. Next time, highlight 6 quotations and ask the student to choose the best 2. Eventually, provide the text without highlighting and see if the student can identify evidence independently (perhaps with more time).

---

Known Limitations

1. The adaptation is based on general research about the learner profile, not on the individual student. Dyslexia manifests differently in different students — some struggle primarily with decoding, others with reading speed, others with working memory. The teacher's knowledge of the specific student is essential for refining the adaptation. If the student's dyslexia primarily affects spelling rather than reading, the adaptations should be different.

2. Differentiation by learning style is explicitly excluded. This skill does not adapt tasks based on "visual," "auditory," or "kinaesthetic" preferences — the evidence does not support this approach (Pashler et al., 2008). Adaptations are based on researched barriers associated with specific learning needs, not on preferences.

3. Adapted tasks can inadvertently signal low expectations. If a student consistently receives "different" work, they may internalise the message that they are less capable. The implementation notes address this, but the teacher must be vigilant about framing adaptations as access support (like glasses for someone who needs them), not as reduced expectations. The goal is equity — the same learning, differently accessed — not a lower track.

Critical Thinking Task Designer

What This Skill Does

Takes a curriculum topic and produces a structured critical thinking task that embeds disciplinary thinking within subject content — with the intellectual resources students need to do the thinking scaffolded explicitly. Critical thinking is domain-contingent: you cannot think critically about something you know too little about, and what counts as good thinking in history is not the same as what counts as good thinking in science or ethics. This skill operationalises that insight. It does not produce generic "higher order thinking" tasks. It produces tasks where the thinking demand is specific to the discipline, the knowledge prerequisites are checked before the task is designed, and the criteria for good thinking are stated explicitly so students know what they are aiming for. The skill draws on Bailin et al.'s intellectual resources framework: a critical thinker needs background knowledge, operational knowledge of good thinking in the domain, knowledge of critical concepts, effective heuristics, and habits of mind. All five must be present or scaffolded for the task to actually develop critical thinking rather than merely demand it. AI is particularly valuable here because designing a good critical thinking task requires simultaneously knowing the content domain, the disciplinary thinking standards, the knowledge prerequisites, and the assessment logic — a combination that is rare in any single educator and that most lesson planning processes skip entirely. This skill also functions as teacher professional development: the act of specifying what good thinking looks like in your subject is itself a significant pedagogical insight that most teachers have never been asked to articulate.

Evidence Foundation

The foundational insight is from Willingham (2007): critical thinking skills are not transferable in the way general skills are. A student who thinks critically in history may think naively in biology, because the standards for good thinking are discipline-specific. Teaching generic critical thinking skills — inference, analysis, evaluation — without grounding them in specific disciplinary content produces students who can name thinking moves but cannot execute them meaningfully. The implication: critical thinking must be developed through disciplines, not alongside them.

Bailin et al. (1999) provide the most useful operational framework. They define the critical thinker through five intellectual resources that must be present: background knowledge (you cannot think critically about something you do not know enough about), operational knowledge of what good thinking looks like in this domain, knowledge of critical concepts (evidence, argument, assumption, perspective), effective heuristics (thinking moves that work in this domain), and habits of mind (intellectual humility, tolerance for ambiguity, commitment to good reasoning). Generic critical thinking instruction typically provides only the third category — concepts — and assumes the others will follow. They do not.

McPeck (1981) argued that critical thinking is entirely domain-specific — there are no general thinking skills, only disciplinary ones. Ennis (1989) countered that some thinking skills transfer across domains. The domain-specificity debate has a pragmatic resolution for curriculum designers: some thinking skills are more transferable than others (identifying assumptions, considering alternative perspectives) but all require domain knowledge to execute meaningfully. The design implication is that critical thinking tasks should be embedded in specific content, with the domain knowledge explicitly checked or provided, and with the thinking standards stated in discipline-specific terms.

Bailin & Siegel (2003) extended this framework within philosophy of education, arguing that critical thinking is not a skill at all but a quality of reasoning that is constituted by the intellectual resources the thinker brings to bear. This reframing is important for task design: the goal is not to "teach critical thinking" as a skill but to ensure that students have the intellectual resources needed to think well about a specific topic, and then to create tasks that require those resources to be deployed.

Paul & Elder (2006) contributed the concept of disciplinary thinking standards — the specific criteria by which reasoning is judged within a discipline. In science, good thinking requires testable hypotheses, controlled variables, and evidence-based conclusions. In history, good thinking requires source evaluation, corroboration, and contextualisation. In ethics, good thinking requires identifying stakeholders, articulating principles, and considering consequences. These standards are not interchangeable, and a task that does not make the relevant standards explicit leaves students guessing at what "good thinking" means.

Hattie (2009) found that teaching thinking skills has a moderate effect size (d = 0.62), but with enormous variation depending on implementation. The evidence suggests that thinking skills instruction is effective when it is embedded in content, when the thinking standards are made explicit, and when students practise applying them to specific problems — precisely the conditions this skill is designed to create. Generic thinking skills instruction divorced from content produces much weaker effects.

Maton's (2013) semantic wave concept is relevant here: effective knowledge-building requires moving between abstract principles and concrete cases. A critical thinking task that stays only at the abstract level ("analyse the argument") without grounding it in specific content produces shallow thinking. A task that stays only at the concrete level ("describe what happened") without requiring students to apply analytical frameworks produces description, not thinking. A well-designed task moves students across the semantic wave: from concrete case to abstract principle and back to concrete application.

Perkins & Salomon (1989) demonstrated that transfer of cognitive skills is not automatic — it requires deliberate bridging. Students do not spontaneously apply thinking skills learned in one domain to another. This reinforces the design principle that critical thinking must be developed within each discipline separately, with explicit attention to which thinking moves are discipline-specific and which might transfer with deliberate teaching.

Input Schema

The educator must provide:

• Curriculum topic: The subject, unit, or concept students are currently studying. e.g. "Year 9 History — causes of World War One" / "Year 10 Biology — natural selection and evolution" / "Year 8 Ethics — the trolley problem and moral reasoning"
• Learner stage: Age range or year group. e.g. "14-15 years" / "Year 9" / "Band D"
• Discipline or subject: The subject area. e.g. "History" / "Biology" / "Ethics" / "Geography"

Optional (injected by context engine if available):

• Knowledge baseline: What students already know about this topic
• Existing learning targets: Any LTs this task should connect to
• Thinking focus: A specific thinking move the teacher wants to emphasise

Prompt

You are an expert in critical thinking pedagogy, with deep knowledge of Bailin et al.'s (1999) intellectual resources framework, Willingham's (2007) research on domain-specificity, McPeck's (1981) and Ennis's (1989) work on critical thinking and subject specificity, Paul & Elder's (2006) disciplinary thinking standards, Hattie's (2009) evidence on thinking skills instruction, and Maton's (2013) semantic wave concept. You understand that critical thinking is domain-contingent: it requires sufficient domain knowledge, discipline-specific thinking standards, and all five intellectual resources (background knowledge, operational knowledge, critical concepts, heuristics, habits of mind) to be present or scaffolded.

Your task is to design a structured critical thinking task for the following curriculum input.

**Curriculum topic:** {{curriculum_topic}}
**Learner stage:** {{learner_stage}}
**Discipline or subject:** {{discipline_or_subject}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Knowledge baseline:** {{knowledge_baseline}} — if not provided, state what the minimum knowledge base would need to be and flag this as an assumption.
**Existing learning targets:** {{existing_learning_targets}} — if provided, connect the task to these LTs.
**Thinking focus:** {{thinking_focus}} — if provided, use this as the primary thinking demand. If not provided, select the most productive thinking demand for this topic and discipline.

## Process

Follow these seven steps precisely. Each step produces a named section in the output.

**Step 1 — Knowledge Prerequisite Check.**
Before designing the task, assess whether students have sufficient domain knowledge for critical thinking to be meaningful. If a student cannot accurately describe the phenomenon they are asked to analyse, they cannot analyse it critically — they will produce opinion dressed as reasoning. State explicitly: what is the minimum knowledge base required for this task? If the knowledge baseline provided suggests students are not there yet, flag this and recommend what to establish first. Do not proceed to task design until this question is answered. If no knowledge baseline is provided, state the assumed prerequisites and flag this as an assumption the teacher must verify.

**Step 2 — Identify the Disciplinary Thinking Standard.**
What does good thinking look like specifically in this discipline and for this topic? Avoid generic standards. "Evaluates evidence" is generic. "In history: identifies the provenance of a source, explains how provenance affects reliability, and corroborates the claim across at least two independent sources" is disciplinary. State the standard in terms that a student could understand and a teacher could use to assess. Reference the discipline's conventions for valid reasoning — what counts as evidence, what counts as a strong argument, what distinguishes expert thinking from novice thinking in this field.

**Step 3 — Select the Thinking Demand.**
Choose one primary thinking move for this task. Options include but are not limited to: evaluating competing claims or interpretations, identifying assumptions underlying a position, analysing a case through multiple lenses, constructing and defending a reasoned argument, identifying the limits of a given explanation, or considering how context or perspective changes interpretation. One thinking demand, done well, produces more genuine critical thinking than five thinking demands done superficially. State the selected thinking demand and explain why it is the most productive choice for this topic and learner stage.

**Step 4 — Design the Task.**
Write the context or scenario students will think about. It should be concrete, discipline-specific, and genuinely open — there should be better and worse answers but not a single correct answer. Write the specific question or prompt. It should require the selected thinking move to answer well. It should not be answerable by recall alone. The task should move students across Maton's (2013) semantic wave: from concrete case to abstract principle and back to concrete application.

**Step 5 — Scaffold the Intellectual Resources.**
For each of Bailin et al.'s (1999) five intellectual resources, state what students need and how it will be provided or activated:

1. **Background knowledge:** What do students need to know, and is it already established or does it need to be provided in the task materials? If it needs to be provided, specify exactly what to include.
2. **Operational knowledge:** What does good thinking look like here, and has the teacher modelled it? If not, how should the standard be shared with students before they begin?
3. **Critical concepts:** Which concepts (evidence, assumption, perspective, causation — discipline-specific) are required, and do students have working definitions? If not, provide the definitions.
4. **Heuristics:** What thinking moves or protocols will help students approach this task? Provide 1-2 specific, actionable heuristics — not generic advice ("think carefully") but concrete moves ("ask: what would change my mind about this?").
5. **Habits of mind:** Which dispositions (intellectual humility, tolerance for ambiguity, willingness to revise) does this task cultivate, and how will the classroom environment support them? Note: these are dispositional — they cannot be assessed through this task but can be cultivated by it.

**Step 6 — Write the Assessment Guidance.**
State 3-4 criteria for quality thinking in student responses. These criteria assess the reasoning process, not the conclusion. A student who reaches an unconventional conclusion through rigorous reasoning should score higher than a student who reaches a conventional conclusion through poor reasoning. Make the criteria specific to this task and this thinking demand. Use the rubric scale Emerging / Developing / Competent / Extending if level descriptors are provided. These criteria become the input for the Coherent Rubric Logic Builder if a formal rubric is needed.

**Step 7 — Write the Teacher Development Notes.**
What does designing and running this task reveal about disciplinary thinking in this subject? What would a teacher who ran this task learn about what good thinking looks like in their discipline? Frame this as professional insight, not classroom management advice. This section is the reason the skill has value beyond individual lesson planning — it builds the teacher's own critical thinking about their subject.

Return your output in this exact format:

## Critical Thinking Task: [Topic]

**Discipline:** [Subject]
**Learner stage:** [Age/year]
**Thinking demand:** [Selected thinking move]
**Connected LTs:** [If provided; otherwise "None specified"]

### 1. Knowledge Prerequisite Check

**Minimum knowledge required:**
[List the specific knowledge elements students must have before this task is meaningful]

**Status:** [Sufficient / Insufficient / Assumed — based on knowledge baseline provided]

[If insufficient: what must be established first, and how]

### 2. Disciplinary Thinking Standard

**In [discipline], good thinking about [topic] requires:**
[State the standard in discipline-specific terms — what counts as evidence, what counts as a strong argument, what distinguishes expert from novice reasoning]

### 3. Thinking Demand

**Primary thinking move:** [Selected demand]
**Why this demand:** [Why it is the most productive choice for this topic and learner stage]

### 4. The Task

**Context:**
[The scenario or case students will think about — concrete, discipline-specific, genuinely open]

**Task materials provided to students:**
[Any documents, data, sources, or information students need — specify exactly what is given]

**The question:**
[The specific prompt — requiring the selected thinking move, not answerable by recall alone]

### 5. Intellectual Resources

| Resource | What students need | How it is provided or activated |
|---|---|---|
| Background knowledge | [Specific knowledge] | [Already established / provided in task materials / needs pre-teaching] |
| Operational knowledge | [What good thinking looks like here] | [How the standard is shared — teacher modelling, exemplar, explicit statement] |
| Critical concepts | [Discipline-specific concepts required] | [Working definitions provided / already established / needs explicit teaching] |
| Heuristics | [Specific thinking moves] | [Stated to students as protocol / modelled by teacher] |
| Habits of mind | [Dispositions cultivated] | [How the classroom environment supports them] |

### 6. Assessment Guidance

**Criteria for quality thinking** (assess reasoning process, not conclusion):

1. [Criterion 1 — specific to this task]
2. [Criterion 2]
3. [Criterion 3]
4. [Criterion 4 — if needed]

**What distinguishes Competent from Extending:**
[The specific quality of reasoning that separates meeting the standard from genuinely exceeding it]

### 7. Teacher Development Notes

[What this task reveals about disciplinary thinking in this subject — framed as professional insight]

**Self-check before returning output:** Verify that (a) the knowledge prerequisite check is specific and honest — not assumed to be sufficient without evidence, (b) the disciplinary thinking standard is genuinely discipline-specific, not generic, (c) the task is not answerable by recall alone, (d) all five intellectual resources are addressed with specific content, not generic placeholders, (e) the assessment criteria assess reasoning quality, not conclusion correctness, (f) the teacher development notes contain genuine disciplinary insight, and (g) the rubric scale uses Emerging / Developing / Competent / Extending wherever level descriptors appear.

Example Output

Scenario: Curriculum topic: "Year 9 History — causes of World War One" / Learner stage: "14-15 years" / Discipline or subject: "History" / Knowledge baseline: "Students have studied the alliance system (Triple Alliance and Triple Entente), the assassination of Archduke Franz Ferdinand and the July Crisis, imperial competition between European powers, and the growth of nationalism in the late 19th century" / Thinking focus: "Evaluating competing historical interpretations"

---

Critical Thinking Task: Causes of World War One

Discipline: History Learner stage: 14-15 years (Year 9) Thinking demand: Evaluating competing historical interpretations Connected LTs: None specified

1. Knowledge Prerequisite Check

Minimum knowledge required:

• The alliance system: which countries were in the Triple Alliance and Triple Entente, and what the alliances committed them to
• The assassination of Archduke Franz Ferdinand: what happened, when, by whom, and the immediate political consequences (Austria-Hungary's ultimatum to Serbia)
• The July Crisis: the sequence of events between the assassination (28 June 1914) and the outbreak of general war (early August 1914) — specifically, which country mobilised when and why
• Imperial competition: what European powers were competing for (colonies, trade routes, prestige), and at least two specific examples of imperial rivalry
• Nationalism: how nationalist sentiment contributed to tensions — at minimum, the difference between unifying nationalism (e.g. Pan-Slavism) and state nationalism (e.g. French revanchism)
• At least a basic awareness that historians disagree about WHO or WHAT was most responsible for the war — the idea of historiographical debate, even if students do not yet know the specific positions

Status: Sufficient — the knowledge baseline covers all required elements. The one area to verify is whether students understand that historians actively disagree about causes, or whether they have been taught "the causes of WWI" as a settled list. If the latter, a brief introduction to historiographical debate is needed before the task.

2. Disciplinary Thinking Standard

In history, good thinking about the causes of WWI requires:

• Source evaluation: Assessing who wrote or said something, when, in what context, and for what purpose — and explaining how these factors affect the reliability and usefulness of the source for a specific claim. A historian citing a German government memo from July 1914 must consider that the memo was written during a diplomatic crisis and may reflect strategic positioning rather than genuine intent.
• Corroboration: Not relying on a single source or a single historian's interpretation. Claims about causation must be checked against multiple sources. If Fischer argues Germany deliberately planned war, the evidence he uses must be examined — and competing evidence from other historians must be considered.
• Distinguishing causation types: Proximate causes (the assassination, the July Crisis) are not the same as structural causes (alliance system, imperial competition, nationalism). Good historical thinking names which type of cause is being discussed and explains why the distinction matters.
• Appropriate qualification: Historical claims about causation should be qualified — "the evidence suggests" rather than "this proves." Good historical thinking acknowledges what the evidence can support and what it cannot.

3. Thinking Demand

Primary thinking move: Evaluating competing historical interpretations Why this demand: At 14-15, students are ready to move beyond "listing causes" (recall) to engaging with the idea that historians disagree about which causes mattered most and why. This is the transition from treating history as a fixed narrative to treating it as an ongoing argument — the fundamental disciplinary shift. Evaluating competing interpretations requires students to assess the quality of reasoning, not just the content of claims, which is the core of historical critical thinking. It also requires them to hold two positions in mind simultaneously and make a reasoned judgment between them — a cognitive demand that is achievable at this age with appropriate scaffolding.

4. The Task

Context: In 1961, the German historian Fritz Fischer published a book arguing that Germany bore primary responsibility for World War One — that Germany's leaders deliberately pursued a policy of aggression, using the assassination of Franz Ferdinand as a pretext to launch a war they had been planning. This was explosive. For decades, the accepted view had been that all major powers stumbled into war through a combination of alliances, miscalculation, and escalation — that no single country was primarily responsible. Fischer's argument divided historians and remains debated to this day.

Task materials provided to students:

Source A — The Fischer Thesis (summarised): Fischer argued that Germany's leaders, particularly Chancellor Bethmann-Hollweg, pursued a "September Programme" of expansionist war aims. He pointed to evidence that Germany gave Austria-Hungary a "blank cheque" of unconditional support, knowing this would likely lead to a wider war. Fischer used German government documents from 1914 to argue that the decision for war was deliberate, not accidental.

Source B — The Revisionist Position (summarised): Historians including Christopher Clark (The Sleepwalkers, 2012) have argued that all major powers share responsibility. Clark described the outbreak of war as a collective failure: Austria-Hungary's reckless ultimatum, Russia's premature mobilisation, France's encouragement of Russia, Britain's ambiguous signals — each power made decisions that escalated the crisis. Clark argued that focusing on German guilt distorts a picture in which multiple actors made catastrophic choices.

Source C — Primary evidence: Telegram from Kaiser Wilhelm II to Tsar Nicholas II, 29 July 1914: "I am exerting my utmost influence to induce the Austro-Hungarian Government to deal straightly to arrive at a satisfactory understanding with you. I confidently hope that you will help me in my efforts to smooth over difficulties that may still arise. Your very sincere and devoted friend and cousin, Willy."

Source D — Primary evidence: Extract from the minutes of the German Imperial War Council, 8 December 1912 (discovered by Fischer): "His Majesty the Kaiser said that... if Russia mobilised, there would be no alternative but war. He further said the Fleet must prepare for war against England... The Chief of the General Staff said that in his opinion war was inevitable — the sooner, the better."

The question: Based on the evidence you have studied and the sources provided, which interpretation — Fischer's argument for German responsibility or the revisionist argument for shared responsibility — is better supported by the available evidence? In your answer:

• State which interpretation you find more convincing and why
• Use specific evidence from the sources AND from your own knowledge to support your position
• Explain what evidence the OTHER side uses that you find hardest to dismiss
• Identify at least one thing the evidence CANNOT tell us — a limit of what we can know about the causes of WWI

5. Intellectual Resources

Resource	What students need	How it is provided or activated
Background knowledge	Alliance system, July Crisis sequence, imperial competition, nationalism — sufficient to evaluate whether the source evidence is consistent with broader patterns	Already established through prior teaching. Teacher should run a brief retrieval activity (5 minutes) at the start of the lesson: "List the alliance commitments and the July Crisis timeline from memory." This activates the knowledge and reveals gaps before the task begins.
Operational knowledge	What a good historical argument looks like — evidence-based, qualified, acknowledges counter-evidence, distinguishes between types of cause	Teacher models this before the task using a different example: take a simpler historical debate (e.g. "Was the Roman Empire's fall caused by barbarian invasion or internal decay?") and demonstrate what a strong evaluation of two positions looks like. 10 minutes. Students see the standard before they are asked to meet it.
Critical concepts	Historical causation: proximate causes (events that triggered the war) vs structural causes (conditions that made war possible). Provenance: who produced a source, when, and why — and how this affects what we can use it for. Corroboration: checking a claim against multiple sources.	Teacher provides working definitions on the board or a handout. Key question: "When we say 'cause' in history, do we mean 'the thing that triggered the event' or 'the condition that made the event possible'? Fischer is arguing about BOTH. Which type of cause does each piece of evidence support?"
Heuristics	"What would change my mind?" — After forming an initial position, students ask: what evidence would I need to see to switch to the other interpretation? This forces genuine engagement with the counter-argument rather than dismissal. "What can this evidence NOT tell us?" — For each source, students identify its limits. The Kaiser's telegram shows what he said to the Tsar, not what he actually intended. The War Council minutes show what was discussed, not what was decided.	Stated explicitly as two thinking protocols on the board. Teacher walks through one example of each before students begin: "Source C — the Kaiser's telegram — sounds conciliatory. But what can it NOT tell us? It cannot tell us whether Wilhelm was sincere or performing diplomacy for the record. That matters for how much weight we give it."
Habits of mind	Intellectual humility: the willingness to say "I'm not sure" or "the evidence is not conclusive." Tolerance for ambiguity: the ability to hold two plausible interpretations in mind without prematurely resolving the tension. Willingness to revise: changing your mind if the evidence warrants it.	Teacher sets the tone explicitly: "In history, saying 'I'm not sure which interpretation is better supported' is not a weakness — it's a sign of mature thinking, as long as you explain WHY you're not sure. I am more impressed by honest uncertainty than by false confidence." Establish that the task rewards quality of reasoning, not strength of conviction.

6. Assessment Guidance

Criteria for quality thinking (assess reasoning process, not conclusion):

1. Use of specific historical evidence. Does the student cite specific evidence from the sources AND from their own knowledge? Vague references ("there is evidence that Germany wanted war") score lower than specific ones ("the War Council minutes from December 1912 show the Kaiser and his Chief of Staff discussing the inevitability of war — Fischer uses this to argue the decision was premeditated"). A student citing Fischer's position is not evidence — the evidence Fischer uses is.

2. Quality of engagement with the counter-argument. Does the student genuinely engage with the strongest version of the opposing interpretation, or do they dismiss it superficially? A student who writes "The revisionist argument is wrong because Germany started it" has not engaged. A student who writes "Clark's argument that all powers share responsibility is supported by Russia's premature mobilisation — this is the hardest evidence for Fischer's thesis to account for, because it suggests the escalation was not solely Germany's doing" has engaged meaningfully.

3. Appropriate qualification of claims. Does the student distinguish between what the evidence supports and what it does not? Claims like "this proves Germany was responsible" are over-stated — the evidence is not conclusive. Claims like "the War Council minutes suggest that at least some German leaders considered war desirable as early as 1912, though the minutes do not prove that this preference became policy" are appropriately qualified.

4. Disciplinary accuracy of historical reasoning. Does the student distinguish between proximate and structural causes? Between what a source says and what it means? Between what the evidence can and cannot support? Students who treat all evidence as equally weighted, or who do not consider provenance, are reasoning below the disciplinary standard.

What distinguishes Competent from Extending: A Competent response evaluates both interpretations using specific evidence, engages with the counter-argument, and qualifies claims appropriately. An Extending response does all of this AND identifies a genuine limitation of the evidence base — something that NEITHER interpretation can fully explain given the available sources — and articulates what further evidence would be needed to resolve the debate. This represents the shift from evaluating within given frameworks to evaluating the frameworks themselves.

7. Teacher Development Notes

What this task reveals about disciplinary thinking in history:

Designing this task forces the teacher to confront a question most history teachers have never been asked explicitly: what is the difference between knowing about the causes of WWI and thinking historically about the causes of WWI? The answer is not obvious. A student who can list all four MAIN causes (alliances, imperialism, militarism, nationalism) and describe the July Crisis in detail has significant knowledge — but they may have no idea how to evaluate whether one cause matters more than another, or how to assess the quality of a historian's argument. The knowledge is necessary but not sufficient for thinking.

The disciplinary thinking standard — source evaluation, corroboration, distinguishing causation types, appropriate qualification — is not something most history curricula make explicit. Students are told to "evaluate sources" but rarely taught what evaluation means in history specifically: that provenance matters because it determines what a source can be used to support, that corroboration matters because no single source is sufficient for a causal claim, and that qualification matters because historical evidence is always incomplete. Making these standards explicit is itself a professional development exercise: the teacher must articulate what they do implicitly when they reason historically, which requires them to examine their own disciplinary thinking.

The Fischer debate is particularly revealing because it shows that historical knowledge is not settled. Students who have been taught "the causes of WWI" as a definitive list may find it disorienting that historians disagree fundamentally about the same events. This disorientation is productive — it is the moment where students begin to understand that history is an argument, not a narrative. But it requires the teacher to be comfortable with ambiguity, which in turn requires the teacher to understand the historiography well enough to guide students through genuine uncertainty rather than retreating to false certainty.

The "what would change my mind?" heuristic is worth noting as a transferable move. While the disciplinary thinking standards are history-specific, the habit of asking "what evidence would make me revise my position?" is a metacognitive move that transfers across disciplines. A teacher who runs this task in history and then asks their science or ethics colleagues to identify the equivalent heuristic in their disciplines is doing precisely the kind of cross-disciplinary professional learning that builds a school-wide culture of critical thinking — one discipline at a time, with the standards made explicit in each.

---

Known Limitations

1. Critical thinking tasks are only as good as the knowledge base that precedes them. This skill includes a prerequisite check, but it cannot verify what students actually know — only what the teacher reports. A task designed for students who "know the alliance system" will fail if they know it only superficially. The teacher must be honest about knowledge sufficiency.

2. The domain-specificity of thinking standards means this skill requires the teacher to have sufficient disciplinary understanding to validate the output. The skill will produce a disciplinary thinking standard — but if that standard does not accurately reflect how experts actually reason in this domain, the task will develop pseudo-critical thinking rather than genuine disciplinary reasoning. The teacher is the quality check.

3. One task does not develop critical thinking. Critical thinking develops through repeated practice of disciplinary thinking across a sustained curriculum. A single well-designed task is evidence of what the curriculum should be doing consistently, not a standalone intervention. If this skill is used once per unit and the rest of the teaching is recall-based, critical thinking will not develop.

4. The habits of mind component (intellectual humility, tolerance for ambiguity) is dispositional and cannot be developed through a single task or assessed through a rubric. The task can cultivate these habits if the classroom culture supports them — but the classroom culture is outside the scope of this skill. If the assessment environment is controlling or high-stakes, students will perform reasoning for the rubric rather than engage in genuine thinking.

5. Transfer between domains should not be assumed. Running this skill in history does not mean students will think critically in science. Each discipline requires its own thinking standards to be made explicit. The appropriate response is to use this skill consistently across multiple subjects, not to assume that critical thinking developed in one domain transfers automatically.

Formative Assessment Technique Selector

What This Skill Does

Selects the most appropriate formative assessment technique for a specific learning moment — during instruction, after guided practice, at the end of a lesson, or between lessons — and provides a complete implementation guide including how to interpret the responses and what to do next based on what the data shows. Unlike generic lists of formative assessment ideas, this skill matches the technique to the specific moment, purpose, and practical constraints. A technique that works brilliantly at the end of a lesson (exit ticket) is useless during a teacher explanation; a technique that works for checking factual recall (mini-whiteboards) is inappropriate for checking deep understanding (which requires explanation, not single answers). AI is specifically valuable here because selecting the right technique requires matching the assessment purpose (what am I checking?), the timing (when in the lesson?), the response format (do I need quick data from everyone, or deep data from a few?), and the practical constraints (class size, resources, time).

Evidence Foundation

Black & Wiliam (1998) established formative assessment as one of the highest-leverage interventions in education (effect size 0.4–0.7), but crucially defined it by its function, not its form: assessment is formative only when the evidence is used to adapt teaching. Giving an exit ticket and not reading it until the next day is not formative assessment — it's delayed summative assessment. Wiliam (2011) operationalised formative assessment into five key strategies: clarifying intentions, engineering discussions, providing feedback, activating students as resources for each other, and activating students as owners of their learning. Leahy et al. (2005) translated these strategies into practical classroom techniques, emphasising that formative assessment must be embedded in instruction — not bolted on. Heritage (2010) distinguished between planned formative assessment (designed into the lesson in advance) and interactive formative assessment (responsive, in-the-moment), arguing that both are necessary but serve different purposes. Wiliam & Leahy (2015) provided comprehensive implementation guidance, emphasising that the best techniques collect evidence from ALL students, not just volunteers.

Input Schema

The teacher must provide:

• Learning moment: When in the sequence. e.g. "During my explanation of how to add fractions with unlike denominators — I need to check before moving on" / "After students have completed 5 practice problems independently" / "End of the lesson — I need to know who's got it and who hasn't" / "Start of next lesson — checking retention from yesterday"
• What to assess: The specific thing to check. e.g. "Can students identify the common denominator?" / "Do students understand why the character made that decision?" / "Can students apply the formula to a new context?"
• Student level: Year group. e.g. "Year 7"

Optional (injected by context engine if available):

• Class size: Number of students
• Subject area: The curriculum subject
• Student profiles: EAL, confidence levels, specific needs
• Available resources: Equipment available

Prompt

You are an expert in formative assessment and responsive teaching, with deep knowledge of Black & Wiliam's (1998) research on assessment for learning, Wiliam's (2011) five key strategies, Leahy et al.'s (2005) practical techniques, and Heritage's (2010) distinction between planned and interactive formative assessment. You understand that formative assessment is defined by its FUNCTION (using evidence to adapt teaching), not its FORM (any particular technique) — and that the right technique depends on the specific learning moment, the type of understanding being assessed, and the practical constraints.

Your task is to select formative assessment techniques for:

**Learning moment:** {{learning_moment}}
**What to assess:** {{what_to_assess}}
**Student level:** {{student_level}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Class size:** {{class_size}} — if not provided, assume 25–30 students.
**Subject area:** {{subject_area}} — if not provided, infer from the assessment focus.
**Student profiles:** {{student_profiles}} — if not provided, assume a mixed-ability class with some EAL students and some reluctant participants.
**Available resources:** {{available_resources}} — if not provided, assume mini-whiteboards are available but no technology.

Apply these evidence-based principles:

1. **Every student, not just volunteers (Wiliam, 2011):**
   - The technique must collect evidence from ALL students, not just those who raise their hands.
   - "Hands up" is NOT formative assessment — it only tells you about the 5–6 students who volunteer. It tells you nothing about the other 20–25.
   - Techniques that collect data from every student: mini-whiteboards, exit tickets, finger voting, all-student response systems, think-pair-share with reporting.

2. **Match technique to purpose (Heritage, 2010):**
   - **Checking factual knowledge:** Quick-response techniques — mini-whiteboards, finger voting, true/false cards. Speed matters; depth doesn't.
   - **Checking understanding:** Explanation-based techniques — think-pair-share, exit tickets with reasoning, hinge questions with diagnostic distractors. Depth matters; speed is secondary.
   - **Checking application:** Task-based techniques — practice problems with monitoring, worked examples with a gap, mini-tasks. Observation of process matters.
   - **Checking misconceptions:** Diagnostic techniques — hinge questions with misconception-targeted distractors, deliberate error identification tasks.

3. **Match technique to timing (Leahy et al., 2005):**
   - **During instruction (real-time):** Must be fast (under 60 seconds), non-disruptive, and provide immediate data. Mini-whiteboards, finger voting, traffic lights.
   - **After guided practice:** Must reveal whether students can apply independently. Quick practice problem, show-me task, partner explanation.
   - **End of lesson:** Must capture what students are leaving with. Exit ticket (3–5 minutes), summary task, one-sentence explanation.
   - **Between lessons:** Must check retention and inform the next lesson. Retrieval quiz at start of next lesson, homework analysis.

4. **Interpret and act (Black & Wiliam, 1998):**
   - The technique is only formative if the teacher USES the data to adapt teaching.
   - For each technique, provide: what response patterns mean, and what to DO based on each pattern.
   - 80%+ correct → proceed. 50–80% → brief re-teach. Below 50% → significant reteach needed.

5. **Avoid pseudo-formative assessment:**
   - "Thumbs up if you understand" is NOT formative assessment — students cannot accurately self-assess understanding in the moment, and social pressure ensures most thumbs go up.
   - "Any questions?" is NOT formative assessment — students who don't understand often don't know what to ask.
   - The technique must require students to DEMONSTRATE understanding, not just claim it.

Return your output in this exact format:

## Formative Assessment: [What's being assessed]

**Moment:** [When in the lesson]
**Assessing:** [Specific understanding/skill]
**For:** [Student level]

### Recommended Techniques

For each technique (2–3, ranked):

**Technique [N]: [Name]**
**How it works:** [Step-by-step implementation]
**Time required:** [Minutes]
**Why this technique for this moment:** [Specific rationale]
**What to look for:** [Specific responses that indicate understanding vs. confusion]

### Response Interpretation Guide

**If 80%+ demonstrate understanding:** [What to do]
**If 50–80% demonstrate understanding:** [What to do]
**If below 50%:** [What to do]
**If you see [specific common error]:** [What it means and what to do]

### Common Mistakes with These Techniques

[2–3 ways the technique can fail and how to prevent it]

**Self-check before returning output:** Verify that (a) the recommended technique collects data from ALL students, (b) the technique is matched to the specific learning moment and purpose, (c) the implementation guide is specific enough for a teacher to use immediately, (d) the response interpretation includes specific actions, and (e) no recommended technique relies on volunteers or self-report.

Example Output

Scenario: Learning moment: "I've just finished explaining how to calculate the area of a triangle (base × height ÷ 2). I've shown two worked examples. Before I set practice problems, I need to check whether students understand the method well enough to try independently." / What to assess: "Can students correctly calculate the area of a triangle when given the base and height?" / Student level: "Year 7" / Subject area: "Mathematics" / Available resources: "Mini-whiteboards available"

---

Formative Assessment: Area of a Triangle Calculation

Moment: After teacher explanation and worked examples, before independent practice Assessing: Can students correctly calculate the area of a triangle when given base and height? For: Year 7 Mathematics

Recommended Techniques

Technique 1: Mini-Whiteboard Show-Me (Recommended) How it works:

1. Display a triangle with base = 8cm, height = 5cm on the board.
2. Say: "Calculate the area of this triangle. Show your working on your whiteboard. You have 60 seconds."
3. After 60 seconds: "3, 2, 1, show me." All students hold up whiteboards simultaneously.
4. Scan the room. Look for the correct answer (20cm²) and the working.
5. If time allows, repeat with a second triangle (different numbers) to confirm.

Time required: 2–3 minutes (including the follow-up problem) Why this technique for this moment: This is a procedural check — can students apply a formula correctly? Mini-whiteboards are ideal because they collect data from every student simultaneously, they show WORKING (not just answers), and they provide instant feedback. The "3, 2, 1, show me" protocol prevents students from copying — everyone reveals at the same time. What to look for:

• Correct answer with correct working (20cm²: 8 × 5 = 40, 40 ÷ 2 = 20) → student is ready for practice
• Answer of 40cm² → student multiplied but forgot to divide by 2
• Answer of 13cm² → student added instead of multiplied (8 + 5 = 13)
• Blank whiteboard or random number → student doesn't know where to start — didn't understand the worked examples

Technique 2: Finger Voting on a Hinge Question How it works:

1. Display: "A triangle has a base of 6cm and a height of 4cm. What is the area?" A) 10cm² B) 12cm² C) 24cm² D) 20cm²
2. Say: "Think about this. Do NOT call out. When I say go, hold up fingers: 1 for A, 2 for B, 3 for C, 4 for D."
3. "3, 2, 1, show me." Scan the room.

Time required: 1–2 minutes Why this technique for this moment: Faster than whiteboards. The distractors are diagnostic: A (10) = added instead of multiplied; B (12) = correct; C (24) = multiplied but didn't divide by 2; D (20) = common plausible wrong answer. Each wrong answer reveals a specific error. What to look for:

• Fingers showing 2 (B, correct: 12cm²) → student understands the method
• Fingers showing 3 (C, 24cm²) → student multiplied correctly but didn't halve — most common error
• Fingers showing 1 (A, 10cm²) → student added base and height instead of multiplying

Technique 3: Partner Explain (for checking deeper understanding) How it works:

1. "Turn to your partner. Person A: explain to Person B how to calculate the area of a triangle. Don't just say the answer — explain the METHOD. Person B: listen and check — are they correct?"
2. After 60 seconds: "Person B — put your hand up if Person A explained it correctly."
3. Cold-call 2–3 Person B students: "What did your partner say?"

Time required: 3–4 minutes Why this technique for this moment: Use this if you want to check whether students can EXPLAIN the method, not just apply it. Explaining to a peer reveals deeper understanding than calculating an answer. However, it's slower and provides less precise data than whiteboards. What to look for: Can students articulate "multiply the base by the height, then divide by 2" — or do they say "I just did 8 times 5 divided by 2" without understanding why? Listen for whether they can explain the ÷ 2 step: "because a triangle is half of a rectangle."

Response Interpretation Guide

If 80%+ get the correct answer (Technique 1 or 2): Proceed to independent practice. Students are ready. Set 6–8 problems with increasing difficulty (including triangles where the height is not a vertical line from the base — this is the next common difficulty).

If 50–80% get the correct answer: Identify the most common error. If most errors are "forgot to ÷ 2" (answer of 40 or 24): brief 2-minute re-teach. "I can see many of you multiplied correctly but forgot one step. Look — a triangle is HALF of a rectangle. So the area is HALF of base times height. Watch me do one more." Re-do one example emphasising the ÷ 2, then re-check with whiteboards.

If below 50% get the correct answer: Do NOT proceed to practice — students will practise the wrong method and embed errors. Return to the explanation. "I can see many of you are unsure. That's fine — this is tricky. Let me show you again from the start." Re-teach using a different representation (e.g., cut a rectangle in half diagonally to show that the triangle is half the rectangle). Then re-check.

If you see "added instead of multiplied" (answer of 13 or 10): This student has confused the area formula with the perimeter concept. They need clarification: "Area is the space INSIDE the shape — we multiply to find how many square centimetres fit inside. Perimeter is the distance AROUND the outside — that's when we add."

Common Mistakes with These Techniques

1. Allowing students to show whiteboards at different times. If students reveal answers one by one, later students copy earlier ones. The "3, 2, 1, show me" simultaneous reveal is essential — it prevents copying and gives you an honest snapshot of the whole class.

2. Only looking at the answer, not the working. A student who writes "20cm²" has the right answer, but did they calculate 8 × 5 ÷ 2, or did they guess? Looking at the working distinguishes genuine understanding from lucky guesses. On whiteboards, require working to be shown.

3. Asking "Does everyone understand?" after the check. If the whiteboard data shows 85% correct, proceed. Don't undermine the data by then asking a question that will always get "yes." Trust the evidence over self-report.

---

Known Limitations

1. Quick formative assessment techniques (whiteboards, finger voting) are best suited for checking procedural knowledge and factual recall. Deeper understanding — why the formula works, when to use it, how it connects to other concepts — requires more time-intensive assessment methods (explanation tasks, extended problems, discussion). This skill recommends the right technique for the moment, but some learning objectives require assessment methods that cannot be completed in 2 minutes.

2. The response interpretation thresholds (80%, 50–80%, below 50%) are guidelines, not rules. A class that is 78% correct may be ready to proceed if the 22% who got it wrong all made the same correctable error. A class that is 82% correct may need to pause if the errors suggest a deep misconception. The teacher's professional judgement must interpret the data in context.

3. Formative assessment only works if the teacher is prepared to adapt. The technique provides data, but the value is in the response. A teacher who checks whiteboards, sees 60% correct, and proceeds anyway because they need to "get through the lesson" has conducted a formative check but not formative assessment. The willingness to adapt the lesson in response to data is the essential ingredient that no technique can provide.

Gap Analysis from Student Work

What This Skill Does

Analyses a student work sample against assessment criteria, identifies specific gaps (not just "needs improvement" but exactly WHAT is missing and WHY), classifies each gap by type (conceptual misunderstanding, procedural error, or communication/presentation issue), and generates targeted next teaching steps — specific actions the teacher can take to close each gap, rather than generic advice like "practise more." The output also includes a feedback script showing how to communicate the analysis to the student in a way that promotes improvement. AI is specifically valuable here because effective gap analysis requires simultaneously comparing the work to the criteria, diagnosing the type of gap (which determines the remedy), identifying strengths (which maintain motivation), and designing a targeted next step (which requires pedagogical knowledge) — a multi-layered analysis that takes significant time and expertise to do well.

Evidence Foundation

Sadler (1989) established that formative assessment depends on three conditions: the student (and teacher) must understand the goal (what quality looks like), assess the current position (where the work is relative to the goal), and take action to close the gap. Gap analysis operationalises the second condition — systematically identifying where the work falls short and why. Hattie & Timperley (2007) demonstrated that effective feedback must address three questions: "Where am I going?" (the goal), "How am I going?" (current performance relative to the goal), and "Where to next?" (specific actions to close the gap). Most teacher feedback addresses only the first two; the third — specific next steps — is where learning happens. Black & Wiliam (1998) showed that formative assessment is only effective when the information gathered is used to adapt teaching — gap analysis without targeted action is diagnostic without being therapeutic. Heritage (2010) emphasised the importance of classifying gaps: a conceptual gap (the student doesn't understand the underlying idea) requires different intervention from a procedural gap (the student understands but makes errors in execution) or a communication gap (the student understands and can do it but can't express it). Wiliam (2011) argued that the most powerful feedback gives the student a specific, actionable next step rather than a judgement.

Input Schema

The teacher must provide:

• Student work description: What the student produced. e.g. "Student wrote: 'The writer uses a metaphor to describe the storm. This is effective because it makes the reader interested.'" / "Student's working: 3/4 + 2/5 = 5/9" / Paste of actual student text
• Assessment criteria: What it should be assessed against. e.g. "The student should identify language techniques, use quotations, and explain the effect on the reader" / "The student should find a common denominator before adding fractions"
• Learning objective: What the student was meant to learn. e.g. "Analyse how writers use language to create effects" / "Add fractions with unlike denominators"

Optional (injected by context engine if available):

• Student level: Year group
• Student profile: Prior attainment, strengths, known gaps
• Subject area: The subject
• Task description: The task the student was completing

Prompt

You are an expert in formative assessment and diagnostic analysis of student work, with deep knowledge of Hattie & Timperley's (2007) feedback framework, Sadler's (1989) model of formative assessment, and Heritage's (2010) gap classification system. You understand that identifying gaps is only useful if each gap is classified (conceptual, procedural, or communication), because the classification determines the correct teaching response.

Your task is to analyse this student work:

**Student work:** {{student_work_description}}
**Assessment criteria:** {{assessment_criteria}}
**Learning objective:** {{learning_objective}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Student level:** {{student_level}} — if not provided, infer from the work and criteria.
**Student profile:** {{student_profile}} — if not provided, base analysis on the evidence in the work itself.
**Subject area:** {{subject_area}} — if not provided, infer from the work and criteria.
**Task description:** {{task_description}} — if not provided, infer from the criteria and objective.

Apply these evidence-based principles:

1. **Identify strengths first (Hattie & Timperley, 2007):**
   - Before identifying gaps, identify what the student does WELL.
   - Strengths must be specific and evidence-based — not generic praise ("good effort") but specific acknowledgment ("correctly identifies the metaphor and selects a relevant quotation").
   - Strengths provide the foundation for improvement — the next step often involves extending a strength.

2. **Classify each gap (Heritage, 2010):**
   - **Conceptual gap:** The student doesn't understand the underlying idea. ("The student adds numerators and denominators directly — they don't understand that fractions with different denominators represent different-sized pieces.")
   - **Procedural gap:** The student understands the concept but makes errors in execution. ("The student finds a common denominator but then forgets to adjust the numerators.")
   - **Communication gap:** The student understands and can do it but can't express it adequately. ("The student can explain the effect verbally but writes 'This is effective' without elaboration.")
   - Classification matters because: conceptual gaps need re-teaching, procedural gaps need targeted practice, and communication gaps need modelling of how to express understanding.

3. **Generate targeted next steps (Hattie & Timperley, 2007; Wiliam, 2011):**
   - Each gap must have a SPECIFIC next teaching step — not "practise more" but "work through 3 examples where the common denominator is given, focusing only on adjusting the numerators."
   - Next steps should be achievable — one or two steps forward, not a complete re-teach.
   - Prioritise: if multiple gaps exist, identify which ONE should be addressed first (usually the most foundational — a conceptual gap before a procedural one).

4. **Write a feedback script that promotes action (Sadler, 1989):**
   - Feedback should be specific, forward-looking, and actionable.
   - Structure: strength acknowledgment → gap identification → specific next step.
   - Avoid: grades or scores (which end learning), vague praise ("good work"), vague criticism ("needs more detail"), and overwhelming lists of everything wrong.

Return your output in this exact format:

## Gap Analysis: [Brief description]

**Student work:** [Brief summary]
**Assessed against:** [Criteria summary]
**Learning objective:** [Objective]

### Strengths

[Specific, evidence-based strengths — what the student does well]

### Gaps Identified

For each gap:
**Gap [N]: [Brief label]**
- **Type:** Conceptual / Procedural / Communication
- **Evidence:** [What in the student's work reveals this gap]
- **What it means:** [What the student does or doesn't understand]
- **Next teaching step:** [Specific, actionable step to close this gap]

### Priority

[Which gap to address first and why]

### Feedback Script

[How to communicate this to the student — specific wording the teacher could use, structured as: acknowledge strength → identify the key gap → provide the next step]

**Self-check before returning output:** Verify that (a) strengths are specific and evidence-based, (b) each gap is classified as conceptual, procedural, or communication, (c) each gap has a specific next teaching step (not "practise more"), (d) gaps are prioritised, (e) the feedback script is forward-looking and actionable, and (f) the feedback doesn't overwhelm — it identifies the most important next step, not every possible improvement.

Example Output

Scenario: Student work: "The writer uses a metaphor 'the sea was a raging beast' to describe the storm. This is a good metaphor because it compares the sea to a beast. The effect on the reader is that they find it effective and want to read on." / Assessment criteria: "Identify language techniques, embed quotations, and analyse the effect on the reader — explaining what the reader thinks, feels, or pictures" / Learning objective: "Analyse how writers use language to create effects" / Student level: "Year 8" / Subject area: "English"

---

Gap Analysis: Language Analysis — Metaphor

Student work: "The writer uses a metaphor 'the sea was a raging beast' to describe the storm. This is a good metaphor because it compares the sea to a beast. The effect on the reader is that they find it effective and want to read on." Assessed against: Identify language techniques, embed quotations, analyse reader effect Learning objective: Analyse how writers use language to create effects

Strengths

1. Correctly identifies the technique. The student names "metaphor" accurately — they can recognise the technique in context. This is a necessary foundation for analysis.
2. Selects a relevant quotation. The quotation "the sea was a raging beast" is well-chosen — it's a clear, analysable example of the technique. The student demonstrates good evidence selection.
3. Embeds the quotation correctly. The quotation is integrated into the sentence with quotation marks, not dropped in as a separate line. This shows awareness of academic writing conventions.
4. Attempts to discuss effect on the reader. The student has the right structure — technique → quotation → effect — which shows they understand WHAT they should be doing, even though the analysis itself is underdeveloped.

Gaps Identified

Gap 1: Generic effect analysis — "effective and want to read on"

• Type: Communication gap (likely — see note below)
• Evidence: The student writes "they find it effective and want to read on." This is a generic claim applied to the metaphor without specifying the ACTUAL effect. "Want to read on" could be said about any technique in any text — it carries no analytical content.
• What it means: The student may understand that the metaphor creates a sense of power and danger (communication gap — they know but can't express it) OR they may not understand what "effect on the reader" means in specific terms (conceptual gap). Diagnostic check needed: ask the student "What does this metaphor make you picture? What feeling does it create?" If they can answer verbally, it's a communication gap. If they can't, it's conceptual.
• Next teaching step: Model the difference between generic and specific effect analysis. Show the student: Generic: "This is effective." Specific: "The word 'raging' suggests the sea is violent and out of control, like an angry animal. The reader pictures enormous, dangerous waves and feels the power and unpredictability of the storm." Practise with 2–3 more quotations: "What does this make you SEE? What does this make you FEEL?"

Gap 2: Explaining the comparison rather than analysing it

• Type: Conceptual gap
• Evidence: "This is a good metaphor because it compares the sea to a beast." The student is DESCRIBING what the metaphor does (compares two things) rather than ANALYSING what it achieves (creates a sense of danger, wildness, power). The student is stuck at identification — "it compares X to Y" — which is true but is not analysis.
• What it means: The student may not yet understand the difference between identifying a technique (naming it and describing what it does structurally) and analysing a technique (explaining what effect it creates and why). This is a common conceptual gap at Year 8 — students have been taught to find techniques but not to analyse them.
• Next teaching step: Explicitly teach the distinction: "IDENTIFYING a metaphor = saying 'it compares X to Y.' ANALYSING a metaphor = explaining what the comparison makes us THINK or FEEL." Provide a formula: "The word/image [specific word] suggests [what it makes you think/picture] because [reason]. This creates a sense of [emotion/atmosphere] for the reader." Practise this formula with the same quotation, then with a new one.

Gap 3: No analysis of specific word choices within the quotation

• Type: Procedural gap
• Evidence: The student quotes "the sea was a raging beast" as a whole but doesn't zoom in on specific words — "raging" (violence, anger, out of control) and "beast" (animal, dangerous, powerful, inhuman). Effective analysis works at the word level, not just the phrase level.
• Next teaching step: Show the student how to zoom in: "You've got a great quotation. Now let's look at the individual words. What does 'raging' make you think of? What about 'beast'? Why didn't Dickens write 'the sea was a large animal'? What's different about 'raging beast'?" This is a procedural skill — the student needs to learn the PROCESS of zooming into individual words.

Priority

Address Gap 2 first (explaining vs. analysing). This is the conceptual gap — until the student understands the difference between describing a technique and analysing its effect, addressing the other gaps won't help. Once the student can explain what a technique DOES TO THE READER (not just what it IS), gaps 1 and 3 will become much easier to close.

Feedback Script

"Right, let me show you what you've done well and one thing to work on.

Strength: You've done three things really well here. You've correctly identified the metaphor — that's right, it IS a metaphor. You've chosen a great quotation — 'the sea was a raging beast' is full of things to analyse. And you've set out your paragraph in the right structure: technique, quotation, effect. So your foundation is solid.

The gap: Here's the thing I want you to work on. You write 'the effect on the reader is that they find it effective.' But that doesn't tell me anything specific. If I read 'the sea was a raging beast,' what do I actually PICTURE? What do I FEEL? Close your eyes — what does a 'raging beast' look like? [Let the student respond.] Exactly — something powerful, dangerous, out of control. THAT is the effect. The metaphor makes the reader picture the sea as something alive and violent and terrifying. That's analysis.

Next step: Here's your rule for next time. After you write your quotation, ask yourself TWO questions: 'What does this make me SEE?' and 'What does this make me FEEL?' Write the answers to those two questions, and you'll be analysing, not just identifying. Try it now with this quotation — rewrite your second and third sentences using those two questions."

---

Known Limitations

1. The analysis depends on the accuracy and completeness of the student work description. A paraphrased summary of student work may lose important details (specific word choices, errors in reasoning, the structure of the response). Where possible, teachers should provide the actual student text rather than a summary.

2. Diagnostic ambiguity between gap types. Some gaps could be conceptual OR communication — the student might understand but not be able to express it, or might genuinely not understand. The analysis flags this ambiguity where it exists and recommends a diagnostic check (e.g., asking the student to explain verbally), but the teacher must make the final classification based on their knowledge of the student.

3. The analysis is based on a single work sample. One piece of work may not represent the student's typical performance — they may have been tired, rushed, or had an off day. The analysis should be treated as evidence about THIS work, not as a definitive assessment of the student's ability. Multiple work samples over time provide a more reliable picture.

KUD Knowledge Type Mapper

What This Skill Does

Takes a curriculum input — a unit description, competency set, or project brief — and produces a complete Know/Understand/Do chart where every element is labelled with its knowledge type (Hierarchical, Horizontal, or Dispositional) and routed to the correct assessment approach. This is the operational bridge between the Curriculum Knowledge Architecture Designer (which diagnoses what types of knowledge are present) and the assessment skills that build the actual instruments: Coherent Rubric Logic Builder for skills, Dispositional Knowledge Assessment Designer for dispositions. Most KUD charts treat all three columns as equivalent. They are not. A Hierarchical Know element (facts with right/wrong answers) and a Horizontal Know element (contested conceptual knowledge) require fundamentally different assessment — the first can be auto-assessed, the second requires interpretive judgment. A Do that is a skill (assessed by rubric against criteria on a specific task) and a Do that is a disposition (assessed through multi-informant observation over time) cannot be assessed the same way — and routing a disposition to a rubric actively undermines it. This skill makes those distinctions explicit so that every element in the KUD chart has a clear, defensible assessment pathway before teaching begins. AI is specifically valuable here because producing a knowledge-typed KUD chart requires simultaneously applying curriculum epistemology (Bernstein's knowledge structures), assessment design logic (Wiliam's assessability constraints), learning science (Hattie's surface/deep/transfer distinction), and motivation theory (Deci & Ryan's autonomy conditions) — a combination that is rare in any single educator and that most planning processes skip entirely.

Evidence Foundation

Wiggins & McTighe (2005, 2011) established Know/Understand/Do as the canonical unit-level planning architecture within Understanding by Design. The KUD framework distinguishes three types of learning: Know (factual and conceptual knowledge that students should acquire), Understand (enduring understandings — transferable insights that survive beyond the unit), and Do (skills and processes that students should be able to perform). The framework's power lies in forcing curriculum designers to specify all three before designing assessment — backwards design requires knowing what you are assessing before you decide how. However, the original UBD framework treats the three columns as internally homogeneous: all Know elements are treated alike, all Understand elements are treated alike, and all Do elements are treated alike. This is the gap the present skill addresses.

Bernstein (1999, 2000) demonstrated that knowledge is not a single substance. Hierarchical knowledge is coherent, explicitly principled, and cumulative — lower-level concepts must be mastered before higher-level ones are accessible, and there are right and wrong answers that can be tested. Horizontal knowledge is organised as specialised interpretive lenses, each with its own criteria for valid analysis — content can be entered from multiple points, and quality is judged by analytical depth rather than factual correctness. These two types exist within the Know and Understand columns of a KUD chart, and they demand different assessment. A Hierarchical Know element ("The water cycle has five stages: evaporation, condensation, precipitation, runoff, infiltration") can be assessed by a quiz — there is a correct answer. A Horizontal Know element ("Different historians interpret the causes of WWI through different lenses — nationalism, imperialism, militarism, alliance systems") cannot be assessed by a quiz, because knowing the lenses is not the same as being able to apply them analytically. The same distinction applies within the Understand column: a Hierarchical understanding ("Energy is conserved — it transforms between forms but is never created or destroyed") has a canonical formulation, while a Horizontal understanding ("Historical events have multiple causes, and the causes you emphasise depend on the interpretive framework you apply") is inherently perspectival.

Hattie (2009) identified three phases of learning — surface (acquiring new knowledge and skills), deep (integrating and consolidating, making connections), and transfer (applying learning to new contexts and problems). Each phase requires different teaching and different assessment. Surface learning aligns with Hierarchical Know elements — facts, vocabulary, procedures that can be tested directly. Deep learning aligns with the Understand column — making connections, seeing patterns, building conceptual structures. Transfer aligns with the most sophisticated Do elements — applying capability in genuinely new contexts. Hattie & Timperley (2007) extended this by showing that feedback must be calibrated to the learning phase: task-level feedback works for surface learning, process-level feedback works for deep learning, and self-regulation feedback works for transfer. A KUD chart that does not distinguish knowledge types cannot calibrate feedback correctly, because the teacher does not know whether a given element requires task-level or process-level feedback.

Wiliam (2011) argued that assessment is only useful if it generates evidence that can be acted upon — and that the design of assessment must begin from the question "What would count as sufficient evidence that this student has learned this?" This assessability constraint is where the KUD knowledge-typing becomes operationally critical. Some KUD elements are directly assessable through a single task (Hierarchical facts, procedural skills). Some require sustained evidence across multiple tasks (Horizontal analytical capabilities, transferable understandings). And some — dispositional elements — are not assessable through tasks at all, because they exist only in patterns of enacted behaviour over time. A KUD chart that does not make these distinctions produces assessment designs that are either invalid (applying rubric criteria to dispositions) or incomplete (failing to plan for the sustained evidence that horizontal and dispositional elements require).

Deci & Ryan (1985, 2000) established through Self-Determination Theory that assessment contexts experienced as controlling — where the learner perceives judgment rather than support — undermine the autonomy and competence needs that are prerequisites for dispositional development. This creates a hard constraint on assessment routing: any KUD element identified as dispositional MUST be routed to an assessment approach that preserves autonomy — multi-informant observation, coaching-modality feedback, developmental conversation — rather than to a rubric that rates and scores. Routing a disposition to a rubric is not merely inefficient; it is actively harmful, because the evaluative frame suppresses the authentic expression that dispositional development requires.

Input Schema

The educator must provide:

• Curriculum input: Description of the unit, competency, or project. e.g. "Band D Wellbeing unit on emotional regulation and self-care — students explore the science of emotions, develop personal self-regulation strategies, and build habits that support their own wellbeing and the wellbeing of others" / "Year 9 History — causes and consequences of WWI" / "D2R project: Smart Water Systems for a local wetland"
• Learner stage: Age range or year group. e.g. "Band D (ages 12-14)" / "Year 9" / "Grade 7-8"

Optional (injected by context engine if available):

• Knowledge architecture output: The epistemic diagnosis from a prior Curriculum Knowledge Architecture Designer run
• Existing learning targets: Any LTs, competency statements, or standards already written for this unit
• Competency framework: The school's dispositional or competency framework
• Programme purpose: Why this unit or competency matters for the mission

Prompt

You are an expert in curriculum design, knowledge epistemology, and assessment architecture. You combine deep knowledge of Wiggins & McTighe's (2005, 2011) Understanding by Design KUD framework with Bernstein's (1999, 2000) theory of knowledge structures, Hattie's (2009) surface/deep/transfer learning phases, Hattie & Timperley's (2007) feedback calibration, Wiliam's (2011) assessability constraints, and Deci & Ryan's (1985, 2000) Self-Determination Theory.

Your task is to produce a complete, knowledge-typed KUD chart for the following curriculum input — one where every element is classified by knowledge type and routed to the correct assessment approach.

**Curriculum input:** {{curriculum_input}}
**Learner stage:** {{learner_stage}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Knowledge architecture output:** {{knowledge_architecture_output}} — if provided, use the epistemic diagnosis as the starting point for knowledge-typing. If not provided, conduct the knowledge-typing from the curriculum input directly.
**Existing learning targets:** {{existing_learning_targets}} — if provided, run a gap analysis mapping each KUD element to existing LTs.
**Competency framework:** {{competency_framework}} — if provided, use it to identify dispositional elements.
**Programme purpose:** {{programme_purpose}} — if provided, use it to check alignment between the KUD and the school's mission.

## The Three Knowledge Types

You MUST classify every KUD element as one of these three types. The classification determines the assessment route.

**Hierarchical Knowledge (Bernstein, 1999)**
Cumulative, prerequisite-ordered knowledge with right/wrong answers or demonstrable procedures. Assessment can be criterion-referenced and often automated. Surface-level feedback (Hattie & Timperley, 2007) is appropriate.

In a KUD chart, Hierarchical elements appear as:
- **Know:** Facts, definitions, vocabulary, formulae, named processes with canonical descriptions
- **Understand:** Principles, laws, or cause-effect relationships with established correct formulations
- **Do (Skill):** Procedures, techniques, and methods with demonstrable correctness

**Horizontal Knowledge (Bernstein, 1999)**
Perspectival, lens-based knowledge where quality is judged by analytical depth and sophistication, not factual correctness. Assessment requires interpretive judgment and analytical criteria. Process-level feedback (Hattie & Timperley, 2007) is appropriate.

In a KUD chart, Horizontal elements appear as:
- **Know:** Frameworks, perspectives, interpretive lenses, contested concepts
- **Understand:** Insights that depend on analytical stance — enduring understandings that admit multiple valid formulations
- **Do (Skill):** Analysis, interpretation, critical evaluation, argumentation — skills where quality exists on a continuum

**Dispositional Knowledge (extended from Bernstein; Deci & Ryan, 1985, 2000)**
Enacted capacities, orientations, and developing competencies that exist only in patterns of behaviour over time. Cannot be assessed through a single task. Assessment MUST use multi-informant observation and coaching-modality feedback. Self-regulation feedback (Hattie & Timperley, 2007) is appropriate. Rubric-based assessment is PROHIBITED — it undermines the autonomy that dispositional development requires (Deci & Ryan, 1985, 2000).

In a KUD chart, Dispositional elements appear as:
- **Know:** Self-knowledge — awareness of one's own patterns, triggers, strengths
- **Understand:** Developmental self-awareness — understanding that dispositions grow, that context matters, that setbacks are part of development
- **Do (Disposition):** Enacted patterns of behaviour — agency, collaboration, self-regulation, empathy, persistence, regenerative mindset

## Critical Rules

1. **The Do column MUST be split into Skills and Dispositions.** Every Do element must be explicitly classified as either a Skill (assessable by rubric on a task) or a Disposition (assessable by multi-informant observation over time). If this split is not made explicit, the assessment design will fail.

2. **No Dispositional element may be routed to a rubric.** Dispositions are assessed through observation protocols, student self-reflection, and developmental conversation — never through scoring criteria applied to a single task. If a Dispositional element appears to need a rubric, it has been misclassified: either it is actually a Skill, or the assessment approach must change.

3. **Knowledge type determines assessment route.** The routing is not optional — it is a logical consequence of what the knowledge IS:
   - Hierarchical Know/Understand → quiz, test, structured response, auto-assessable
   - Horizontal Know/Understand → analytical task assessed by teacher judgment using criteria
   - Hierarchical Do (Skill) → procedural demonstration, criterion-referenced rubric
   - Horizontal Do (Skill) → performance task, rubric with qualitative descriptors (Emerging → Developing → Competent → Extending)
   - Dispositional Do (Disposition) → multi-informant observation protocol, developmental conversation, student self-reflection (invoke Dispositional Knowledge Assessment Designer)
   - Dispositional Know/Understand → embedded in developmental conversation, not separately tested

4. **Flag compound elements.** If a KUD element contains multiple knowledge types ("Understands how emotions work AND manages their own emotional responses"), flag it for splitting. The first part is Hierarchical/Horizontal Know; the second is Dispositional Do.

5. **Flag unassessable elements.** If a KUD element cannot be assessed validly with any available approach (too vague, too internal, or too context-dependent), flag it and suggest reformulation.

## Process

Step 1: Read the curriculum input carefully. Identify every distinct learning element — what students will know, understand, and be able to do.

Step 2: Classify each element by knowledge type (Hierarchical, Horizontal, or Dispositional). Justify each classification.

Step 3: For every Do element, make the Skill/Disposition split explicit.

Step 4: Assign the assessment route for each element.

Step 5: Identify sequencing constraints — which elements are prerequisites for which? Which can be taught in flexible order? Which dispositions run as continuous threads? For each dispositional Do element identified, ask: is this disposition knowledge-contingent? That is, does it require a sufficient base of hierarchical Know and horizontal Understand elements before it can meaningfully manifest? If yes, identify which specific Know and Understand elements from the KUD chart are prerequisites for this disposition to be operable — not just supportive, but genuinely prerequisite. Include these in the sequencing notes as hard prerequisites for the dispositional element, not just as parallel threads. A disposition that is treated as a continuous thread when it is actually knowledge-contingent will be under-developed because teachers will assume it is developing in parallel rather than recognising that foundational knowledge must come first.

Step 6: If existing LTs are provided, run the gap analysis: map each KUD element to existing LTs, identify gaps (KUD elements with no corresponding LT) and mismatches (LTs that don't align with any KUD element or that are routed to the wrong assessment type).

Step 7: Flag design issues — compound elements, misclassified elements, unassessable elements, and any routing conflicts.

Return your output in this exact format:

## KUD Knowledge Type Map: [Unit/Project Name]

**Curriculum input:** [Summarised]
**Learner stage:** [Age/year/band]
**Programme purpose:** [If provided; otherwise "Not specified"]

### 1. Know

| Element | Knowledge Type | Assessment Route | Justification |
|---|---|---|---|
| [Factual/conceptual element] | Hierarchical / Horizontal / Dispositional | [Route] | [Why this type] |

### 2. Understand

| Element | Knowledge Type | Assessment Route | Justification |
|---|---|---|---|
| [Enduring understanding] | Hierarchical / Horizontal / Dispositional | [Route] | [Why this type] |

### 3. Do — Skills

| Element | Knowledge Type | Assessment Route | Justification |
|---|---|---|---|
| [Skill element] | Hierarchical / Horizontal | [Route — which rubric skill to invoke] | [Why this type] |

### 4. Do — Dispositions

| Element | Knowledge Type | Assessment Route | Justification |
|---|---|---|---|
| [Dispositional element] | Dispositional | Multi-informant observation → Dispositional Knowledge Assessment Designer | [Why this is a disposition, not a skill] |

### 5. Assessment Routing Plan

[For each KUD section, specify:
- Which assessment skill to invoke next
- What inputs that skill needs from this KUD chart
- The assessment mode (auto-assessable / rubric / observation protocol / mixed)
- The feedback type (task-level / process-level / self-regulation)]

### 6. Sequencing Notes

**Hard prerequisites:** [Elements that must be taught before others — hierarchical chains]
**Flexible-order elements:** [Elements that can be taught in any order — horizontal lenses]
**Continuous threads:** [Dispositional elements that develop across the entire unit, not in specific lessons]
**Convergence points:** [Where different knowledge types must come together — e.g. applying hierarchical knowledge through a horizontal lens]

### 7. Gap Analysis

[If existing LTs provided:]

| KUD Element | Matched LT | Match Quality | Notes |
|---|---|---|---|
| [Element] | [LT if matched] | Full / Partial / No match | [What's missing or misaligned] |

**Gaps:** [KUD elements with no corresponding LT]
**Orphan LTs:** [Existing LTs that don't map to any KUD element]
**Routing mismatches:** [LTs where the current assessment approach doesn't match the knowledge type]

[If no existing LTs provided: "No existing LTs provided. To generate LTs from this KUD chart, invoke the Learning Target Authoring Guide with each Do element as input."]

### 8. Design Flags

**Compound elements:** [Elements that contain multiple knowledge types and should be split]
**Misclassified elements:** [Elements that appear to be one type but are actually another — e.g. a "skill" that is actually a disposition]
**Unassessable elements:** [Elements too vague, too internal, or too context-dependent to assess validly — with suggested reformulations]
**Routing conflicts:** [Cases where the assessment route for one element conflicts with another — e.g. a rubric task that would undermine a dispositional goal being developed simultaneously]

**Self-check before returning output:** Verify that (a) every Do element is classified as either Skill or Disposition with no ambiguity, (b) no Dispositional element is routed to a rubric, (c) every element has a knowledge type AND an assessment route, (d) sequencing notes identify ALL prerequisite dependencies, (e) the gap analysis (if applicable) catches genuine misalignments not just superficial mismatches, and (f) design flags are specific and actionable.

Example Output

Read references/worked-examples.md when you need the complete emotional-regulation KUD example, including Know/Understand/Do tables, assessment routing and gap analysis.

Compact excerpt — skill, disposition and private habits:

Element	Knowledge Type	Assessment Route	Justification
Select and apply an appropriate calming technique for a specific situation, explaining why that technique suits that situation	Horizontal	Rubric task — given a scenario, select technique and justify → invoke Coherent Rubric Logic Builder	The selection is Horizontal (multiple valid choices depending on context and reasoning) even though the knowledge of techniques is Hierarchical. Assessment must evaluate the quality of justification, not just whether the "right" technique was chosen.
Self-regulation as an enacted practice — noticing emotional escalation in real time and choosing a response rather than reacting automatically	Dispositional	Multi-informant observation → invoke Dispositional Knowledge Assessment Designer	This is NOT a skill that can be demonstrated on a task. A student can explain self-regulation perfectly on a worksheet and still react impulsively in the corridor. The disposition exists only in patterns of enacted behaviour across contexts and over time. Assessment requires teacher observation, student self-reflection, and developmental conversation. A rubric would assess performance of self-regulation, not development of it.
Wellbeing as habit — maintaining personal routines (sleep, exercise, reflection) that support sustained wellbeing	Dispositional	Student self-reflection and developmental conversation → invoke Dispositional Knowledge Assessment Designer	Personal habits are private and contextual. The school cannot and should not monitor whether a student sleeps well or exercises. Assessment is through the student's own developing self-awareness: can they articulate what works for them, notice when their habits slip, and self-correct? This is self-regulation feedback territory (Hattie & Timperley, 2007), not task feedback.

---

Known Limitations

1. The three knowledge types are a simplification, not an ontology. Real knowledge elements often sit on boundaries between types. A skill like "selecting a calming technique with justification" has Hierarchical components (knowing the techniques), Horizontal components (justifying the selection), and Dispositional components (actually using it). The classification captures the PRIMARY type for assessment routing purposes — it does not claim the element is purely one type.

2. Assessment routing assumes the downstream skills exist. This KUD chart routes elements to the Coherent Rubric Logic Builder and the Dispositional Knowledge Assessment Designer. If neither has been run for these elements, the routing is a plan, not an accomplished fact. The KUD chart tells the teacher WHAT to assess and HOW — the downstream skills produce the actual assessment instruments.

3. The gap analysis depends on the quality of existing LTs. If the existing LTs are poorly written (unobservable verbs, compound constructs, vague language), the gap analysis will flag problems that are really LT writing issues, not KUD mapping issues. In such cases, the recommendation is to invoke the Learning Target Authoring Guide before attempting to resolve the gaps.

4. Dispositional elements are the hardest to assess and the easiest to neglect. The KUD chart makes them visible and routes them to appropriate assessment — but a teacher under time pressure may still default to assessing only the Hierarchical and Horizontal elements because they are easier. The chart cannot solve this problem; it can only make the choice visible.

5. The KUD chart is a planning tool, not a teaching sequence. The sequencing notes provide guidance on prerequisite dependencies and flexible-order elements, but the actual lesson-by-lesson sequence depends on the teacher's judgment about pacing, student needs, and contextual factors that the chart cannot anticipate.

Learning Progression Builder

What This Skill Does

Maps the learning progression from novice to target proficiency for a specific skill domain, identifying the sequential stages of understanding, the prerequisite relationships between them (what must come before what), common stuck points (where students typically stall and why), and diagnostic tasks that reveal which stage a student is currently at. The output is a progression map that teachers can use for three purposes: planning instruction (teaching in the right sequence), formative assessment (diagnosing where a student is), and differentiation (providing the right support for each student's current stage). AI is specifically valuable here because constructing a valid learning progression requires both deep content knowledge (understanding the logical structure of the domain) and pedagogical knowledge (knowing where students actually get stuck, which is not always where the content logic would predict).

Evidence Foundation

Heritage (2008) defined learning progressions as "descriptions of the successively more sophisticated ways of thinking about a topic that can follow one another as children learn." She emphasised that progressions are hypothesised pathways, not rigid tracks — students may skip stages, revisit earlier stages, or take alternative routes. Popham (2007) argued that learning progressions are essential for formative assessment because they provide the "map" that makes it possible to locate a student's current understanding and identify the next step. Without a progression, a teacher knows a student is "struggling" but not WHERE in the learning pathway the difficulty lies. Daro et al. (2011) demonstrated that mathematics learning trajectories — empirically validated progressions — provide the foundation for coherent curriculum, assessment, and instruction. Wilson & Bertenthal (2005) applied learning progressions to science assessment, showing that progression-based assessment is more informative than standards-based assessment because it reveals the developmental pathway, not just whether a binary standard is met. Hattie & Donoghue (2016) showed that different learning strategies are effective at different stages of learning — surface strategies (memorisation, rehearsal) are effective early; deep strategies (elaboration, organisation) are effective later — which means the teaching approach should match the student's position on the progression.

Input Schema

The teacher must provide:

• Target skill: What students should be able to do at the end. e.g. "Solve multi-step equations with the unknown on both sides" / "Write a developed analytical paragraph about a text" / "Design and evaluate a fair scientific experiment"
• Student level: Year group range. e.g. "Year 7–9" / "KS3" / "Middle school"

Optional (injected by context engine if available):

• Subject area: The curriculum subject
• Starting point: Where students begin
• Student profiles: Class data showing current positions
• Curriculum framework: Relevant standards or progression documents

Prompt

You are an expert in learning progressions and curriculum coherence, with deep knowledge of Heritage's (2008) framework for learning progressions, Popham's (2007) work on progression-based assessment, and Hattie & Donoghue's (2016) research on stage-appropriate learning strategies. You understand that learning progressions are hypothesised pathways — they describe the typical developmental sequence but acknowledge that individual students may follow different routes.

Your task is to build a learning progression for:

**Target skill:** {{target_skill}}
**Student level:** {{student_level}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Subject area:** {{subject_area}} — if not provided, infer from the target skill.
**Starting point:** {{starting_point}} — if not provided, identify the typical entry point for students at the beginning of the stated level range.
**Student profiles:** {{student_profiles}} — if not provided, design for a typical class where students are at various points along the progression.
**Curriculum framework:** {{curriculum_framework}} — if not provided, build on general curriculum expectations.

Apply these evidence-based principles:

1. **Identify sequential stages (Heritage, 2008):**
   - Define 5–7 stages from novice to target proficiency.
   - Each stage should describe a qualitatively different level of understanding or capability — not just "more" of the same thing.
   - Each stage should be OBSERVABLE — described in terms of what the student can DO, not what they "understand" internally.
   - Stages should be ordered by typical developmental sequence, acknowledging that some students may not follow this exact order.

2. **Map prerequisite relationships (Daro et al., 2011):**
   - Which stages MUST come before which? (Not just which usually do, but which logically must.)
   - Identify both linear prerequisites (A must come before B) and parallel prerequisites (both C and D must be in place before E).
   - Distinguish hard prerequisites (the stage cannot be attempted without the prior) from soft prerequisites (the stage is easier with the prior but possible without it).

3. **Identify common stuck points (Popham, 2007):**
   - Where do students typically stall? These are the diagnostic priorities.
   - For each stuck point: what does "stuck" look like, and what is usually causing it?
   - Stuck points often occur at transitions between qualitatively different types of thinking (e.g., from procedural to conceptual, from concrete to abstract).

4. **Design diagnostic tasks (Heritage, 2008; Popham, 2007):**
   - For each stage, provide a quick task (2–5 minutes) that reveals whether a student has reached that stage.
   - Diagnostic tasks should be efficient — they test the KEY indicator of each stage, not everything a student at that stage can do.
   - The task should distinguish between adjacent stages — a student at Stage 3 should pass the Stage 3 diagnostic but fail the Stage 4 diagnostic.

5. **Stage-appropriate teaching approaches (Hattie & Donoghue, 2016):**
   - Early stages: surface strategies — explicit instruction, modelling, practice with feedback.
   - Middle stages: deep strategies — elaboration, connection-making, explaining reasoning.
   - Later stages: transfer strategies — application to new contexts, evaluation, independent problem-solving.

Return your output in this exact format:

## Learning Progression: [Target Skill]

**From:** [Starting point]
**To:** [Target proficiency]
**For:** [Student level range]

### Progression Map

For each stage:
**Stage [N]: [Name]**
- **What the student can do:** [Observable indicators]
- **Key shift from previous stage:** [What's qualitatively different]
- **Prerequisite:** [What must be in place first]
- **Diagnostic task:** [Quick task that reveals whether the student is at this stage]

### Prerequisite Diagram

[Visual or textual representation of which stages depend on which]

### Common Stuck Points

For each stuck point:
**Stuck between Stage [X] and Stage [Y]**
- **What "stuck" looks like:** [Observable signs]
- **What usually causes it:** [The underlying difficulty]
- **How to unstick:** [Targeted teaching intervention]

### Teaching Implications

[How the progression should inform instruction — what to teach first, where to invest time, when to use which teaching strategies]

**Self-check before returning output:** Verify that (a) stages are qualitatively distinct, (b) each stage has observable indicators, (c) prerequisite relationships are explicit, (d) diagnostic tasks distinguish between adjacent stages, (e) stuck points are based on common patterns, and (f) the progression represents a developmental pathway, not just a list of topics.

Example Output

Scenario: Target skill: "Write a developed analytical paragraph about a literary text, using embedded quotations and explaining the effect of language on the reader" / Student level: "Year 7–9" / Subject area: "English Literature"

---

Learning Progression: Analytical Writing About Literature

From: Can retell what happens in a text (narrative summary) To: Can write a developed analytical paragraph using embedded quotations and explaining the effect of language choices on the reader For: Year 7–9 English Literature

Progression Map

Stage 1: Retelling

• What the student can do: Summarises what happens in the text in chronological order. "Scrooge is a mean old man who doesn't like Christmas. Then three ghosts visit him and he becomes nice."
• Key shift from previous stage: Entry point — no prior stage required.
• Prerequisite: Basic reading comprehension (can understand the events of the text).
• Diagnostic task: "Tell me what happens in Chapter 1." If the student can retell accurately, they're at Stage 1. If they can ONLY retell (and cannot do Stage 2), they need to progress.

Stage 2: Identifying Features

• What the student can do: Points out language features without analysis. "The writer uses a simile." "There is alliteration in this sentence." Can name techniques when they see them.
• Key shift from previous stage: Moves from "what happens" to "how it's written" — a fundamental shift from content to craft.
• Prerequisite: Stage 1 (must understand the text before analysing it). Must know the names of at least 5 techniques.
• Diagnostic task: "Highlight and label any language techniques you can find in this paragraph." If the student can identify 2+ techniques correctly, they're at Stage 2.

Stage 3: Quotation + Comment

• What the student can do: Selects a quotation and makes a comment about it: "The writer says 'dark and stormy night' which shows it was scary." The comment is relevant but undeveloped — it's a single statement, not an explanation.
• Key shift from previous stage: Moves from just naming techniques to selecting evidence and making a claim about it.
• Prerequisite: Stage 2 (must be able to identify features). Must understand that quotations serve as evidence.
• Diagnostic task: "Choose a quotation from the text and explain what it shows." If the student selects a relevant quotation and makes one relevant comment (even if brief), they're at Stage 3.

Stage 4: Explaining Effect

• What the student can do: Explains what the language does to the reader: "The word 'raging' suggests the sea is violent and out of control, which makes the reader feel afraid for the character." The explanation is specific — it names a feeling, image, or thought the reader experiences.
• Key shift from previous stage: Moves from commenting on what the text says to explaining what the text DOES — from content to effect. This is the critical analytical shift.
• Prerequisite: Stage 3 (must be able to select evidence and comment). Must understand that writers make choices to create effects on readers.
• Diagnostic task: "What effect does the word 'raging' have on the reader? What does it make you think or feel?" If the student can name a specific effect (not "it's effective" or "it makes you want to read on"), they're at Stage 4.

Stage 5: Developed Analytical Paragraph

• What the student can do: Writes a paragraph that follows a coherent structure: point → embedded quotation → analysis of specific words → explanation of effect → link to wider meaning. Each part connects logically to the next.
• Key shift from previous stage: Moves from isolated comments to sustained, structured analysis — multiple sentences that build on each other.
• Prerequisite: Stage 4 (must be able to explain effect). Must understand paragraph structure and cohesion.
• Diagnostic task: "Write one paragraph analysing how the writer creates tension in this extract." If the paragraph has a clear point, embedded evidence, specific word-level analysis, and explained effect, the student is at Stage 5.

Stage 6: Evaluative and Comparative Analysis (Target)

• What the student can do: Writes analytical paragraphs that consider multiple possible interpretations, compare techniques, evaluate which is most effective, and connect analysis to the writer's wider purpose and themes. Uses evaluative and comparative language: "While the metaphor creates visual power, it is the sentence structure that truly conveys the character's distress."
• Key shift from previous stage: Moves from single-interpretation analysis to evaluative, comparative, multi-layered analysis — the student considers WHY the writer made this choice over alternatives.
• Prerequisite: Stage 5 (must be able to write a developed analytical paragraph). Must have sufficient knowledge of the wider text and context.
• Diagnostic task: "Analyse how the writer creates [effect] in this extract. Consider at least two techniques and evaluate which is more effective." If the student compares techniques and evaluates with reasoning, they're at Stage 6.

Prerequisite Diagram

Stage 1 (Retelling)
    ↓
Stage 2 (Identifying Features) ← [requires: knowledge of technique names]
    ↓
Stage 3 (Quotation + Comment)
    ↓
Stage 4 (Explaining Effect) ← [critical shift: content → effect]
    ↓
Stage 5 (Developed Paragraph) ← [requires: paragraph structure skills]
    ↓
Stage 6 (Evaluative Analysis) ← [requires: wider text knowledge]

Linear progression with lateral prerequisites at Stages 2, 5, and 6.

Common Stuck Points

Stuck between Stage 2 and Stage 3: Can identify techniques but can't select evidence

• What "stuck" looks like: The student highlights and labels techniques enthusiastically but freezes when asked to "choose a quotation and explain it." They can spot features but can't select the BEST one for analysis.
• What usually causes it: The student has been trained to identify ALL techniques (feature-spotting) but not to evaluate which quotation is worth writing about. They see techniques as things to find, not things to analyse.
• How to unstick: Teach quotation selection: "Here are three quotations. Which one gives you the most to write about? Why?" Practise choosing between quotations before asking students to find their own.

Stuck between Stage 3 and Stage 4: Can comment but can't explain effect (THE MOST COMMON STUCK POINT)

• What "stuck" looks like: The student writes "This shows that..." and makes a relevant comment, but the comment describes the content, not the effect. "The metaphor shows the sea is dangerous" (content) vs. "The metaphor makes the reader feel afraid" (effect). Or the student defaults to generic effect claims: "This is effective and makes the reader want to read on."
• What usually causes it: The student doesn't distinguish between what the text SAYS and what it DOES. They have been taught to find meaning in the text but not to consider the reader's experience. This is a conceptual barrier — they need to understand that analysis is about the EFFECT of language, not just its meaning.
• How to unstick: Two-question technique: after every quotation, ask "What does this make you SEE?" and "What does this make you FEEL?" These questions force attention toward the reader's experience. Model extensively: show the difference between content comments and effect analysis side by side.

Stuck between Stage 5 and Stage 6: Can write a paragraph but can't evaluate or compare

• What "stuck" looks like: The student writes competent PEEL paragraphs that analyse one technique at a time, but cannot compare techniques, consider alternative interpretations, or evaluate relative effectiveness. Their paragraphs are solid but independent — they don't build toward a larger analytical argument.
• What usually causes it: The student has mastered the procedural skill (writing an analytical paragraph) but hasn't developed the evaluative thinking to go beyond it. Comparison and evaluation are higher-order skills that require the student to hold multiple analyses in mind simultaneously.
• How to unstick: Teach comparison explicitly: "You've written about the metaphor. Now write about the sentence structure. Now — which technique is MORE effective at creating tension, and why?" Provide a comparison frame: "While [technique 1] creates [effect], [technique 2] is arguably more effective because..."

Teaching Implications

1. Invest the most time at the Stage 3→4 transition. This is where most students stall and where the critical analytical shift occurs. A student who can explain effect is ready for rapid progress through Stages 5 and 6; a student who can't explain effect will plateau regardless of how much they write.

2. Don't skip stages. A student who can't yet identify techniques (Stage 2) should not be asked to write an analytical paragraph (Stage 5). The stages build on each other — skipping to the end task without building the foundation produces overwhelm, not learning.

3. Use diagnostic tasks to differentiate. In any Year 8 class, students will be spread across Stages 2–5. Use the diagnostic tasks to locate each student, then provide stage-appropriate instruction: Stage 2 students practise identification, Stage 3 students practise quotation selection, Stage 4 students practise effect explanation, Stage 5 students practise paragraph construction.

4. Match teaching approach to stage. Stages 1–3: explicit instruction, modelling, guided practice (surface strategies). Stage 4: think-alouds, worked examples, peer discussion (deep strategies). Stages 5–6: independent practice, self-assessment against criteria, comparative writing tasks (transfer strategies).

---

Known Limitations

1. Learning progressions are hypothesised pathways, not fixed tracks. Individual students may skip stages, regress temporarily, or develop skills in a different order. The progression describes the TYPICAL developmental sequence — the teacher must use professional judgment when students don't follow the expected path.

2. The progression describes skill development in ONE domain. A student may be at Stage 5 for poetry analysis but Stage 3 for prose analysis, because the underlying texts present different challenges. Progressions are domain-specific — the teacher should assess each domain separately.

3. Diagnostic tasks provide a snapshot, not a comprehensive assessment. A student who passes the Stage 4 diagnostic task on one occasion may not consistently perform at Stage 4. The diagnostic locates the student's approximate position — ongoing formative assessment provides the more complete picture.

Project Brief Designer (PBL)

What This Skill Does

Designs a complete project brief for project-based learning — including a driving question, real-world connection, structured milestones, explicit instruction points, and assessment criteria — that ensures students learn substantive content THROUGH the project rather than simply producing a product. The critical design principle is that effective PBL combines authentic, open-ended inquiry with structured teaching: the project provides the motivation and context; explicit instruction provides the knowledge and skills students need to succeed. The output is a ready-to-use project brief that a teacher can hand to students, plus a teacher-facing implementation guide that maps where direct instruction, formative assessment, and scaffolding are needed. AI is specifically valuable here because designing effective PBL requires simultaneously balancing authenticity (making the project feel real), rigour (ensuring substantive learning happens), structure (building in milestones that prevent drift), and differentiation (making the project accessible to all learners) — a multi-dimensional design challenge that takes significant expertise and time.

Evidence Foundation

Barron & Darling-Hammond (2008) reviewed evidence on inquiry-based learning and identified the design features that distinguish effective projects from activities that are engaging but educationally shallow: effective PBL connects to meaningful real-world problems, requires disciplinary thinking (not just information gathering), includes structured milestones, and incorporates explicit instruction at the points where students need new knowledge or skills. They found that PBL is most effective when it supplements, not replaces, direct instruction — the project provides a context that makes instruction meaningful, and instruction provides the tools that make the project possible. Krajcik & Shin (2014) identified five key features of effective PBL: a driving question (authentic, open-ended, anchored in real-world issues), situated inquiry (investigation embedded in meaningful context), collaboration, learning technologies, and tangible artefacts. They emphasised that the driving question is the design centrepiece — it must be genuinely open (not a question with a predetermined answer), connected to students' lives, and rich enough to sustain extended investigation. Larmer, Mergendoller & Boss (2015) from the Buck Institute for Education established the "Gold Standard PBL" framework with seven essential design elements: a challenging problem or question, sustained inquiry, authenticity, student voice and choice, reflection, critique and revision, and a public product. Thomas (2000) reviewed PBL research and found positive effects on content knowledge and problem-solving but cautioned that poorly designed projects can be time-consuming without producing proportionate learning — structure and explicit instruction are the differentiating factors. Hmelo-Silver, Duncan & Chinn (2007) demonstrated that scaffolded inquiry outperforms unscaffolded inquiry — students need structured support, not just open-ended tasks.

Input Schema

The teacher must provide:

• Project topic: What the project is about. e.g. "Water quality in our local river" / "Designing a sustainable city" / "The impact of the Industrial Revolution on working people" / "Creating a campaign to reduce food waste in our school"
• Learning objectives: What students should learn. e.g. "Understand the causes and effects of water pollution, apply scientific testing methods, communicate findings to an audience" / "Analyse primary and secondary sources about working conditions, construct historical arguments using evidence"
• Student level: Year group. e.g. "Year 8" / "Year 10"
• Project duration: How long. e.g. "6 lessons over 3 weeks" / "Half a term (12 lessons)"

Optional (injected by context engine if available):

• Subject area: The curriculum subject(s)
• Real-world connection: Specific context the teacher wants
• Student profiles: Class data, interests, needs
• Available resources: Technology, materials, community links
• Curriculum framework: Standards to address

Prompt

You are an expert in project-based learning design, with deep knowledge of Barron & Darling-Hammond's (2008) research on inquiry-based learning, Krajcik & Shin's (2014) five key features of effective PBL, and Larmer, Mergendoller & Boss's (2015) Gold Standard PBL framework. You understand that effective PBL is NOT simply "do a project" — it is a carefully designed learning experience where authentic inquiry and explicit instruction work together so that students learn substantive content THROUGH the project.

CRITICAL DESIGN PRINCIPLE: PBL effects are strongest when projects include explicit instruction, not instead of it (Barron & Darling-Hammond, 2008; Hmelo-Silver et al., 2007). Every project brief you design must include specific points where the teacher provides direct instruction, modelling, or scaffolding. A project without structured teaching is an activity, not PBL.

Your task is to design a project brief for:

**Project topic:** {{project_topic}}
**Learning objectives:** {{learning_objectives}}
**Student level:** {{student_level}}
**Project duration:** {{project_duration}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Subject area:** {{subject_area}} — if not provided, infer from the topic and objectives.
**Real-world connection:** {{real_world_connection}} — if not provided, design an authentic connection that makes the project meaningful to students at this level.
**Student profiles:** {{student_profiles}} — if not provided, design for a mixed-ability class.
**Available resources:** {{available_resources}} — if not provided, assume standard classroom resources with internet access but no specialist equipment.
**Curriculum framework:** {{curriculum_framework}} — if not provided, align to the stated learning objectives.

Apply these evidence-based PBL design principles:

1. **Design a driving question (Krajcik & Shin, 2014):**
   - The driving question must be OPEN-ENDED — it should not have a single correct answer.
   - It must be AUTHENTIC — connected to real-world issues, audiences, or problems that matter beyond the classroom.
   - It must be FEASIBLE — students at this level can meaningfully investigate it within the time available.
   - It must REQUIRE the intended learning — students cannot answer the question without developing the knowledge and skills in the learning objectives.
   - Avoid pseudo-questions that have predetermined answers. "How can we reduce water pollution in our local river?" is open. "What are the three types of water pollution?" is not.

2. **Structure milestones with explicit instruction (Barron & Darling-Hammond, 2008; Hmelo-Silver et al., 2007):**
   - Break the project into 3–5 milestones, each with a clear deliverable.
   - At each milestone, identify: what students produce, what they need to know/be able to do, and WHERE EXPLICIT INSTRUCTION HAPPENS.
   - Instruction should be "just in time" — taught when students need it for the next phase of the project, not front-loaded as a lecture block followed by a project block.
   - Each milestone should include a formative check — how does the teacher know students are learning, not just producing?

3. **Ensure authentic audience and purpose (Larmer et al., 2015):**
   - The final product should be FOR someone — a real audience, a genuine purpose.
   - "Present your findings to the class" is weak. "Present your water quality report to the local council environmental officer" is strong.
   - If a real audience isn't available, create a realistic scenario that simulates one.

4. **Build in student voice and choice (Larmer et al., 2015):**
   - Students should have meaningful choices within the project — which aspect to investigate, how to present findings, which evidence to prioritise.
   - Choice should be structured, not unlimited — too much choice overwhelms; too little removes ownership.
   - The learning objectives are non-negotiable; the route to them includes choice.

5. **Design assessment for both process and product (Barron & Darling-Hammond, 2008):**
   - Assess the LEARNING, not just the PRODUCT. A beautiful poster with no substantive content should not score well.
   - Include process assessment: research logs, draft work, peer feedback, reflection.
   - Assessment criteria should be transparent from the start — students should know how they'll be assessed before they begin.
   - Product quality matters, but content understanding matters more.

6. **Include reflection and revision (Larmer et al., 2015):**
   - Build in structured reflection points: "What have you learned so far? What questions do you still have?"
   - Build in revision opportunities: students improve their work based on feedback before final submission.
   - Critique and revision is where much of the learning happens — not in the first draft.

Return your output in this exact format:

## Project Brief: [Project Title]

**Driving Question:** [The open-ended question that drives the project]
**For:** [Student level]
**Duration:** [Project duration]
**Subject:** [Subject area(s)]

### The Scenario

[A compelling, authentic description of the project context — who needs this, why it matters, what students will produce. Written in language appropriate for the student level. This is what students actually read.]

### What You Will Produce

[Clear description of the final product(s) and who the audience is]

### Milestones

For each milestone:
**Milestone [N]: [Name] — [Timing]**
- **What you produce:** [Deliverable]
- **What you'll need to learn:** [Knowledge/skills required]
- **Teacher instruction point:** [What the teacher explicitly teaches at this stage and why]
- **Formative check:** [How the teacher assesses learning at this stage]
- **Student choice:** [Where students have meaningful choice within this milestone]

### Assessment Criteria

**Your project will be assessed on:**
[Clear, transparent criteria — what counts, what doesn't, how much each element is worth]

**Process assessment (how you work):**
[What process elements are assessed — research quality, collaboration, reflection, revision]

**Product assessment (what you produce):**
[What product elements are assessed — content accuracy, argument quality, communication effectiveness]

### Explicit Instruction Map (Teacher Guide)

[For each milestone: what the teacher needs to teach, when, and how. This ensures the project includes genuine instruction, not just facilitation.]

### Differentiation Notes

[How the project can be adapted for different learners — extension challenges, support scaffolds, EAL modifications — while maintaining the same learning objectives]

### Known Design Choices

[Transparent notes about trade-offs in the design — what was prioritised, what was sacrificed, what the teacher should watch for]

**Self-check before returning output:** Verify that (a) the driving question is genuinely open-ended and requires the intended learning, (b) every milestone includes an explicit instruction point, (c) assessment criteria are transparent and prioritise content learning over product aesthetics, (d) student voice and choice are structured within each milestone, (e) the project includes reflection and revision opportunities, and (f) the project could not be completed without substantive learning — it's not possible to produce a good product by copying information without understanding it.

Example Output

Scenario: Project topic: "Water quality in our local river" / Learning objectives: "Understand the causes and indicators of water pollution, apply scientific testing methods (pH, dissolved oxygen, nitrate levels), evaluate evidence to draw conclusions, communicate scientific findings to a non-specialist audience" / Student level: "Year 8" / Project duration: "8 lessons over 4 weeks" / Subject area: "Science" / Real-world connection: "The local river has had reports of increased algae growth — a real local issue"

---

Project Brief: River Health Investigation

Driving Question: How healthy is our local river, and what should be done about it? For: Year 8 Science Duration: 8 lessons over 4 weeks Subject: Science (with elements of Geography and Communication)

The Scenario

In recent months, residents near the River Arden have noticed increasing amounts of green algae growing on the surface of the water, particularly near the Millbrook industrial estate. Some dog walkers have reported their dogs becoming ill after drinking from the river. The local parish council wants to understand what's happening — but they don't have scientific expertise.

They need a team of scientists to:

1. Test the water quality at different points along the river
2. Identify what's causing the problem
3. Explain their findings in plain language
4. Recommend what should be done

Your team has been commissioned to investigate. Your findings will be presented in a scientific report written for the parish council — people who care about the river but don't have a science background. Your report needs to be scientifically accurate AND understandable to non-scientists.

What You Will Produce

Final product: A scientific report (800–1200 words) for the parish council that includes:

• Your water quality findings with data
• A clear explanation of what the data means
• An evidence-based conclusion about the river's health
• Specific recommendations for action

Audience: The parish council (non-specialist adults who need to make decisions based on your findings)

Presentation: Each team will deliver a 5-minute summary of their report to the class, with the teacher role-playing as a parish council member who asks questions.

Milestones

Milestone 1: Understanding Water Quality — Lessons 1–2

• What you produce: A research summary (one page) identifying the key indicators of water quality and what affects them. Include at least 4 indicators and explain what "healthy" and "unhealthy" levels look like for each.
• What you'll need to learn: What water quality means scientifically, the key indicators (pH, dissolved oxygen, nitrate/phosphate levels, biological indicators), what causes pollution in rivers, and what eutrophication is.
• Teacher instruction point: Lesson 1 — direct instruction on water quality indicators. The teacher explains pH, dissolved oxygen, nitrates, and phosphates, using diagrams to show how excess nutrients cause algal blooms (eutrophication). This is essential knowledge that students cannot discover independently within the time available. Lesson 2 — teacher models how to read and interpret water quality data tables, demonstrating what "normal" ranges look like and how to spot concerning readings.
• Formative check: Exit ticket at end of Lesson 2: "A river sample shows pH 6.2, dissolved oxygen 4mg/L, and nitrate levels of 25mg/L. Is this river healthy? Explain your reasoning using at least two indicators." This checks whether students can interpret data, not just recall what indicators exist.
• Student choice: Teams choose which two indicators they will focus their investigation on (all must cover pH and dissolved oxygen; teams choose their additional focus areas based on interest).

Milestone 2: Collecting and Recording Data — Lessons 3–4

• What you produce: A completed data table with water quality measurements from at least 3 sampling points, recorded accurately with units and repeated measurements.
• What you'll need to learn: How to use testing equipment (pH meter, dissolved oxygen probe, nitrate test strips), how to design a fair test (controlling variables when sampling), how to record data accurately in tables.
• Teacher instruction point: Lesson 3 — teacher demonstrates each piece of testing equipment, modelling the correct procedure and common errors. Students practise with prepared samples before going to the river. This is a practical skill that requires explicit modelling — students cannot figure out how to use a dissolved oxygen probe by experimentation. Lesson 4 — brief instruction on fair testing: why we sample at the same depth, take repeated readings, and record conditions. This connects to the broader science skill of controlling variables.
• Formative check: Teacher circulates during data collection, checking: Are readings recorded with correct units? Are students taking repeat measurements? Can students explain WHY they're testing at multiple points along the river? (If students can't explain why, they're following a procedure without understanding the scientific reasoning.)
• Student choice: Teams choose their sampling points along the river (within the approved area) based on their hypothesis about where pollution might be worst. They must justify their choice scientifically.

Milestone 3: Analysing Evidence and Drawing Conclusions — Lessons 5–6

• What you produce: A data analysis document that includes: data presented in appropriate graphs/charts, a written interpretation of what the data shows, and a conclusion about the river's health supported by evidence.
• What you'll need to learn: How to select appropriate graph types for different data, how to identify patterns and anomalies in data, how to write evidence-based conclusions (claim → evidence → reasoning).
• Teacher instruction point: Lesson 5 — teacher models how to analyse water quality data using a sample dataset (NOT the students' own data). Demonstrates: choosing bar charts for comparing sites vs. line graphs for showing change along the river, identifying what the pattern means ("dissolved oxygen decreases as we move downstream from the industrial estate — this suggests..."), and distinguishing between correlation and cause. Lesson 6 — teacher provides a Claim-Evidence-Reasoning (CER) framework and models writing one conclusion paragraph, thinking aloud about how to connect data to a claim.
• Formative check: Each team submits a draft conclusion paragraph (just one) before the end of Lesson 6. The teacher reviews these overnight and provides targeted written feedback on whether the conclusion is supported by the evidence cited. This is the critical learning moment — can students use data to support a scientific argument?
• Student choice: Teams choose how to present their data visually (graph type, layout, which comparisons to highlight). They also choose which findings to emphasise in their conclusion based on what they judge to be most significant.

Milestone 4: Communicating to a Non-Specialist Audience — Lessons 7–8

• What you produce: The final report (800–1200 words) and a 5-minute presentation summary.
• What you'll need to learn: How to communicate scientific findings to non-specialists — translating technical language into accessible explanations without losing accuracy, structuring a report for a decision-making audience.
• Teacher instruction point: Lesson 7 — teacher shows two example paragraphs side by side: one written for scientists ("The elevated NO₃⁻ concentration of 28mg/L suggests anthropogenic eutrophication") and one written for the parish council ("The nitrate levels in the river are higher than they should be. Nitrates act like fertiliser for algae — when levels rise above 10mg/L, algae can grow out of control"). Students identify the differences and practise "translating" one paragraph of their own analysis. This is a specific communication skill that requires modelling.
• Formative check: Peer review in Lesson 7: each team reads another team's draft report and answers three questions: (1) Can you understand the findings without a science background? (2) Is there a clear recommendation? (3) What's one thing that needs more explanation? Teams then revise based on peer feedback before final submission in Lesson 8.
• Student choice: Teams choose the format and structure of their final report (within the word count) and choose how to deliver their presentation (all members speak, one presents while others handle questions, etc.).

Assessment Criteria

Your project will be assessed on:

Process assessment (30% of total mark):

• Research quality (Milestone 1): Did you accurately identify and explain water quality indicators? (10%)
• Data collection (Milestone 2): Was your data recorded accurately, with repeat measurements and correct units? (10%)
• Reflection and revision: Did you improve your report based on peer feedback? What specific changes did you make and why? (10%)

Product assessment (70% of total mark):

• Scientific accuracy (25%): Are your findings correct? Are your conclusions supported by the data? Do you correctly explain the science of water pollution?
• Evidence-based argument (20%): Do you use your data as evidence? Is your reasoning clear? Does your conclusion follow from the evidence?
• Communication to audience (15%): Can a non-specialist understand your report? Have you translated scientific language without losing accuracy?
• Recommendations (10%): Are your recommendations specific, realistic, and based on your findings?

Note: Visual design of the report is NOT assessed. A plain-text report with excellent science scores higher than a beautifully designed report with weak science. Focus your time on understanding and argument, not decoration.

Explicit Instruction Map (Teacher Guide)

Lesson	Explicit Instruction	Purpose	Timing
1	Water quality indicators, eutrophication process	Foundational knowledge students need before investigating	25 minutes direct instruction + 20 minutes guided research
2	Reading and interpreting water quality data tables	Procedural skill: how to make sense of numerical data	15 minutes modelling + 30 minutes practice with sample data
3	Equipment demonstration and fair testing	Practical skill requiring modelling — cannot be self-taught safely	20 minutes demonstration + 25 minutes supervised practice
4	Brief review of variables and sampling design	Connecting practical work to scientific method	10 minutes instruction + field work
5	Data analysis: graph selection, pattern identification	Analytical skill: moving from data to meaning	20 minutes modelling with sample dataset + 25 minutes guided analysis
6	CER framework for scientific conclusions	Writing skill: structuring evidence-based argument	15 minutes modelling + 30 minutes drafting with feedback
7	Communicating science to non-specialists	Communication skill: audience awareness	15 minutes modelling + 30 minutes revision
8	— (student presentations and final submission)	Assessment and celebration	—

Total explicit instruction time: Approximately 120 minutes across 8 lessons. The remaining time is structured inquiry, collaborative work, and formative assessment. This is NOT "teacher talks for 2 lessons then students do a project for 6 lessons" — instruction is distributed throughout, delivered when students need it.

Differentiation Notes

Extension: Students who complete milestones early can investigate additional indicators (biological indicators such as invertebrate diversity), compare their findings to published data for similar rivers, or write a second audience-specific version of their report (e.g., a version for Year 5 students studying rivers in Geography).

Support: Pre-teach water quality vocabulary in a brief session before Lesson 1. Provide a data recording template with units and column headers already filled in. Offer a report structure template with section headings and sentence starters for each paragraph. These scaffolds structure the output without reducing the analytical demand — the student still must interpret the data and draw conclusions.

EAL: Provide a glossary of key scientific terms with everyday-language definitions. Use visual diagrams of the eutrophication process alongside written explanations. Allow students to discuss findings in their first language before writing in English. Provide sentence frames for the CER paragraph: "Our data shows that ___. This suggests ___ because ___."

ADHD: Break each lesson's tasks into shorter segments with clear "done" checkpoints. Provide a visual timeline of the project milestones. During field work, assign a specific role (equipment manager, data recorder) to provide focus and structure.

Known Design Choices

1. The river scenario is based on a real local issue. If the river/issue doesn't match your local context, adapt the scenario — the design principles work with any authentic environmental issue. The key requirement is that the problem is REAL (or realistically simulated), not contrived.

2. The report is prioritised over the presentation. The report carries 70% of the assessment; the presentation is a communication exercise, not a performance assessment. This avoids the validity threat of assessing public speaking confidence instead of scientific understanding (see Assessment Validity Checker).

3. Visual design is explicitly excluded from assessment. This is a deliberate choice to prevent students from investing time in decoration rather than science. It also removes a socioeconomic barrier (access to colour printing, design software) and an equity barrier (students with dyslexia or fine motor difficulties are not penalised).

4. The project requires explicit instruction at every milestone. This is the most important design feature. PBL research consistently shows that projects without structured teaching produce engagement without proportionate learning (Thomas, 2000). If you remove the instruction points, the project becomes an activity — students will produce reports by copying information from websites rather than by understanding the science.

5. Four weeks is the minimum viable duration for this project. Shorter timelines would require cutting data collection (which removes the authentic inquiry) or cutting the revision cycle (which removes a key learning opportunity). If time is limited, reduce the number of sampling points rather than cutting milestones.

---

Known Limitations

1. The quality of PBL depends heavily on the driving question. The generated driving question is designed to be open-ended, authentic, and requiring the intended learning — but the teacher should evaluate whether it genuinely engages their specific students. A question that works in one context may fall flat in another. The teacher may need to adapt the driving question to connect with their students' interests and local context.

2. Real-world connections require local knowledge. The project brief generates a scenario based on the stated topic and real-world connection, but the teacher knows their community, local resources, and potential external partners better than any AI. The generated scenario should be treated as a strong starting point that benefits from local adaptation — replacing generic details with specific local names, places, and issues.

3. PBL is not appropriate for all learning objectives. Some content is better taught through direct instruction, practice, and retrieval — particularly foundational knowledge that students need before they can investigate (Hmelo-Silver et al., 2007). PBL works best for objectives that involve application, analysis, evaluation, and communication — not for objectives that are primarily about acquiring factual knowledge. The teacher should consider whether PBL is the right approach for the stated objectives before using this skill.

4. The explicit instruction map is a guide, not a script. The suggested instruction points indicate WHERE teaching is needed but cannot specify the exact teaching approach that will work for every class. The teacher must use their professional judgement about how much instruction to provide, how to respond to misconceptions that arise during the project, and when to pause the project for additional teaching that wasn't anticipated in the original design.

Scope and Sequence Designer

What This Skill Does

Takes a programme description and developmental band structure and produces a coherent scope and sequence — mapping what gets taught across all bands, in what order, with explicit reasoning for the sequencing decisions. This skill works at any level of education: early childhood through upper secondary, undergraduate, professional development programmes, or any other staged learning architecture. Most scope and sequence documents are lists: topics assigned to year groups without coherent logic for why that topic sits there, what it builds on, or what it prepares students for. This skill produces a structured progression grounded in three knowledge types: hierarchical elements are sequenced by prerequisite logic so foundational knowledge is always in place before the next layer is introduced; horizontal elements are sequenced to build thinking sophistication progressively rather than repeating the same thinking moves at the same level year after year; dispositional elements are mapped as continuous threads with explicit identification of the knowledge prerequisites that must be in place before a disposition can meaningfully develop. When KUD charts, LT types, or a pre-built prerequisite map are supplied, the skill applies them directly; when they are not, it infers prerequisite relationships and flags the confidence of every inference. The result is a programme where every element has a defensible reason for being where it is — and where the epistemic status of each recommendation is explicit. AI is specifically valuable here because coherent programme design requires simultaneously tracking prerequisite chains across years, monitoring knowledge type balance within and across bands, and identifying gaps and overlaps that are invisible when looking at individual units in isolation — a level of systematic cross-referencing that is cognitively demanding and frequently skipped in real curriculum planning.

Evidence Foundation

Bruner (1960) established the foundational principle of the spiral curriculum: key ideas should be revisited across year groups at increasing levels of sophistication, with each encounter building on prior encounters rather than repeating them. A scope and sequence that revisits a topic without increasing the cognitive demand is not a spiral — it is repetition. The spiral principle applies differently to the three knowledge types. Hierarchical knowledge spirals by adding new layers of complexity on top of secured foundations — fractions before algebra, cell biology before genetics. Horizontal knowledge spirals by increasing the sophistication of analytical thinking applied to recurring themes — a student who identifies perspectives at Band A should be evaluating and comparing analytical frameworks at Band D. Dispositional knowledge does not spiral in the same way — it develops continuously through enacted practice across the full programme, though the knowledge that supports dispositional expression deepens at each band.

Schmidt, Wang & McKnight (2005) analysed curriculum coherence across high-performing education systems and found that coherent curricula share three features: focus (fewer topics taught more deeply), rigour (appropriate challenge at each level), and coherence (topics connect logically within and across years). Systems that lack coherence — where topics appear and disappear without progression logic — consistently underperform. Their most significant finding for scope and sequence design is that coherence is not just a vertical property (does Band B build on Band A?) but also a horizontal property (do the elements within Band B connect to each other?). A programme can have perfect vertical sequencing and still lack coherence if the units within each band are isolated from each other.

Duschl, Schweingruber & Shouse (2007) developed the concept of learning progressions: empirically grounded descriptions of how student understanding develops across years, with each level building specifically on the previous one. Their work establishes that progression is not automatic — it requires deliberate curriculum design that matches what is taught to what students are ready to learn. Learning progressions are best documented for hierarchical knowledge domains (mathematics, early reading, some areas of science), where the prerequisite structure is well-researched. For horizontal and dispositional domains, progressions are less empirically established and must be constructed from developmental principles rather than from replicated research on specific learning sequences.

Wiggins & McTighe (2005) applied backwards design to programme-level planning: begin with the intended outcomes at the end of the programme, then work backwards to determine what must be in place at each stage to reach those outcomes. At the scope and sequence level, this means the final band's expectations determine what must be taught in every preceding band — not as direct preparation for a test, but as the knowledge and capability foundations that make the final outcomes achievable.

Bernstein (1999) and Muller (2009) establish the theoretical foundation for knowledge-type-specific sequencing. Hierarchical knowledge has an inherent sequencing logic: concepts must be taught in prerequisite order because later concepts genuinely depend on earlier ones. You cannot teach genetic inheritance before students understand cell division. Horizontal knowledge does not have prerequisite chains in the same way — different analytical lenses can be introduced in various orders — but it does have a sophistication progression: students should move from identifying perspectives to analysing through perspectives to evaluating and synthesising across perspectives. Sequencing horizontal knowledge by increasing analytical demand rather than by prerequisite dependency is one of the key distinctions this skill makes.

Maton (2013) adds the semantic wave concept: effective knowledge-building requires movement between abstract principles and concrete cases across a programme, not just within individual lessons. A scope and sequence that stays at the abstract level throughout produces disconnected theoretical knowledge; one that stays at the concrete level produces experience without conceptual development. Across a programme, the semantic profile should show increasing capacity to move between concrete and abstract — students at early bands work primarily with concrete cases and simple abstractions, while students at later bands should be able to operate fluently at multiple levels of abstraction and move between them deliberately.

Hattie (2009) identified curriculum coherence as a high-effect variable in student achievement. Programmes where students experience learning as a connected, cumulative journey produce better outcomes than programmes where each year feels like a fresh start with new content that does not obviously connect to what came before. This is the practical justification for investing in scope and sequence design: the coherence of the programme is a stronger predictor of student outcomes than the quality of any individual unit within it.

Bransford, Brown & Cocking (2000) and Kolb (1984) establish the experiential readiness principle for dispositional sequencing: students who have practised a disposition encounter the explanation of it as confirmation of lived experience rather than abstraction. Experience should generally precede explanation for social-emotional dispositional capabilities. Flavell (1979) introduces an important exception: for metacognitive and reflective LTs, naming strategies explicitly before practising them improves practice quality. Kirschner, Sweller & Clark (2006) provide the counterpoint that novice learners benefit from explicit instruction before practice in many domains — this skill acknowledges the tension and flags it for teacher discretion wherever it applies.

Input Schema

The educator must provide:

• Subject or programme: Name and brief description. e.g. "Wellbeing, REAL School Budapest" / "Science, Years 7–13" / "Interdisciplinary Humanities, Foundation through Diploma" / "Early Childhood Mathematics, ages 3–6"
• Developmental bands: The band, year group, or stage structure. e.g. "Bands A–D (approximately ages 5–15)" / "Years 7–13" / "Foundation, Intermediate, Advanced, Diploma" / "Nursery, Reception, Year 1, Year 2"
• Intended outcomes: The overarching goals students should reach by the end of the programme. e.g. "Students develop self-regulation, agency, care for others, and the scientific literacy to understand their own wellbeing" / "Students develop scientific reasoning, experimental design capability, and the knowledge base for further study at university level"

Optional (injected by context engine if available):

• Existing units or competencies: Any existing units, competencies, or LTs already in place
• Knowledge architecture output: From curriculum-knowledge-architecture-designer if already run
• Time available: Hours or lessons per week per band
• KUD charts: KUD charts (Know/Understand/Do per band) for the LTs being sequenced. Richest input for prerequisite inference — Know layer reveals T1 content dependencies, Understand layer reveals conceptual scaffold relationships, Do layer reveals T3 vs T1/T2 distinctions. High confidence inference.
• LT types: T1/T2/T3 classification per LT or competency. If not supplied and kud_charts not supplied, skill infers types from LT language.
• Prerequisite map: Pre-built typed prerequisite relationships between LTs, each typed as hard, soft_enabler, or conceptual_accelerator. If supplied, used directly without inference.
• Sequencing principles: Programme-specific sequencing rules that override the skill's default logic where they conflict.

Prompt

You are a curriculum sequencing specialist producing scope and sequence recommendations for competency-based developmental band programmes. You apply three distinct sequencing logics based on content type and prerequisite relationships. You are explicit about the epistemic status of every recommendation — distinguishing hard constraints from soft preferences from teacher professional judgement calls.

---

STEP 0 — INPUT ASSESSMENT AND ROUTING

Assess input state before producing any output:

STATE A — prerequisite_map supplied: Use directly. Highest confidence. Proceed.

STATE B — kud_charts supplied (no prerequisite_map): Infer from KUD charts. High confidence. Proceed.

STATE C — LTs with lt_types supplied (no KUD, no map): Infer from LT content and types. Medium confidence. Proceed with confidence flags.

STATE D — only competency definitions supplied: Low confidence. Produce output with disclaimer: "Prerequisite inference from competency definitions alone is unreliable. Running KUD Chart Author skill first will produce substantially more reliable sequencing." Proceed with low confidence flagged throughout.

STATE E — only subject name and intended outcomes supplied: Decline to produce a scope and sequence. Output: "Insufficient inputs for reliable sequencing. Recommended sequence: (1) Run Learning Target Authoring Guide to produce LTs. (2) Run KUD Chart Author to produce KUD charts. (3) Return here." Do not produce a sequence in State E.

---

STEP 1 — PREREQUISITE MAP

1a. If prerequisite_map supplied: use directly. Skip to Step 2.

1b. Inference from KUD charts (State B):
For each pair of LTs:
- Know layers: if LT A's Know content is directly required by LT B's Know or Understand layer — hard prerequisite.
- Understand layers: if LT A's Understand makes LT B's Understand richer and more portable — conceptual_accelerator.
- Do layers: if LT A is T1 and LT B is T3, and T1 explains why the T3 disposition works — soft_enabler.
- No meaningful dependency — none.
Confidence: high for Know-layer dependencies, medium for Understand-layer inferences.

1c. Inference from LT content and types (State C). Apply by type combination:

T1 → T1: Would a student lacking LT A's content be unable (not just hindered) to access LT B? If yes: hard. If enriching but not gating: soft_enabler. Example: "Understanding the stress response mechanism (T1) is a hard prerequisite for evaluating stress management interventions (T1)."

T1 → T3: T1 content explaining why a T3 disposition works is typically a conceptual_accelerator, not a hard prerequisite. The disposition can develop through practice without explanation — the explanation makes it transferable. Example: "Neuroscience of emotion regulation (T1) is a conceptual accelerator for self-regulation practice (T3)."

T2 → T3: Typically a soft_enabler. Example: "Reflective decision-making (T2) enriches metacognitive self-direction (T3) but does not gate it."

T3 → T3: Usually soft. One T3 disposition rarely makes another logically inaccessible. Example: "Self-awareness (T3) is a soft enabler for empathy (T3)."

T2 → T2: Rarely hard prerequisites — usually parallel. Flag as none unless clear content dependency exists.

Confidence for all State C inferences: medium.

1d. Always include with inferred maps: "These relationships were inferred from LT content and types. Subject expert review is required before treating inferred hard prerequisites as non-negotiable — particularly in mathematics, science, and language acquisition where prerequisite structures are non-obvious from text alone."

---

STEP 2 — THREE SEQUENCING LOGICS

HARD PREREQUISITE LOGIC
Applies to: T1 LTs with hard prerequisites in the map.
Rule: prerequisite LT must appear earlier. Non-negotiable. Violation = PREREQUISITE_VIOLATION error, not a suggestion.
Output: constraint_type: hard in sequencing_constraints.

SOFT SCAFFOLD LOGIC
Applies to: soft_enabler and conceptual_accelerator relationships.
Rule: place enabler/accelerator earlier where possible. When not possible, flag the trade-off in sequencing_rationale.
Output: constraint_type: recommended. Teacher can adjust.

EXPERIENTIAL READINESS LOGIC
Applies to: T3 dispositional LTs.
Default rule: experience of the capability should generally precede T1/T2 content that explains it. Students who have practised a disposition encounter the explanation as confirmation of lived experience, not abstraction. (Bransford, Brown & Cocking (2000). How People Learn. National Academies Press. Kolb (1984). Experiential Learning. Prentice Hall.)

Exception — metacognitive and reflective T3 LTs: For LTs in metacognition and reflection, light conceptual framing before practice may be warranted. Naming metacognitive strategies explicitly before practising them improves practice quality. (Flavell (1979). Metacognition and cognitive monitoring. American Psychologist, 34(10), 906-911.) Flag metacognitive T3 LTs as candidates for concept-first sequencing in sequencing_questions_for_teacher.

Counterargument to acknowledge in output: explicit instruction research (Kirschner, Sweller & Clark (2006). Why minimal guidance during instruction does not work. Educational Psychologist, 41(2), 75-86.) argues novice learners benefit from explicit instruction before practice. Experiential readiness default applies most strongly to social-emotional dispositional capabilities. Do not apply to T1 LTs.

Output: flag T3 ordering as constraint_type: recommended with teacher_discretion_flag.

SEQUENCING PRINCIPLES OVERRIDE
If {{sequencing_principles}} supplied: apply after the three default logics. Where a supplied principle conflicts with a default, the principle wins — but flag the conflict: "Supplied principle [text] overrides the default [logic name] recommendation for [LT name]. If intentional, no action needed. If not, review the supplied principle."

---

STEP 3 — KNOWLEDGE ARCHITECTURE DIAGNOSIS

Before producing the full sequence, identify what types of knowledge are present in this programme. If a knowledge architecture output is provided, use it. If not, conduct a rapid diagnosis: what are the hierarchical elements that have prerequisite chains, what are the horizontal elements that require thinking sophistication to develop, and what are the dispositional elements that develop continuously across the programme? List the major elements under each type. The sequencing logic for each type is fundamentally different and must be treated separately.

---

STEP 4 — FULL SEQUENCE CONSTRUCTION

Apply the three sequencing logics from Step 2 to place every LT or competency in a band. For each placement:
- State the constraint_type: hard or recommended
- State the sequencing_rationale in one sentence: which logic drove the decision
- Flag any teacher_discretion items (all T3 ordering, all soft scaffold decisions)

---

STEP 5 — COHERENCE CHECKS

VERTICAL COHERENCE: For hierarchical elements: is every concept introduced after its prerequisites are secured? For horizontal elements: is analytical sophistication genuinely increasing, or are students doing the same thinking with slightly harder content? For dispositional elements: are development opportunities present throughout? Flag every break.

HORIZONTAL COHERENCE: Within each band: is there appropriate balance between knowledge types? Are units connected or siloed? Would a student finishing this band have the knowledge, thinking, and dispositional development needed to succeed in the next band?

---

STEP 6 — SEQUENCING PRINCIPLES OUTPUT

Write a readable list of principles applied:
- The three default logics, stated plainly
- Any programme-specific principles supplied
- The metacognitive T3 exception
- Invitation: "Review these principles. If any do not match your programme's philosophy or your knowledge of your students, adjust the sequence accordingly."

---

STEP 7 — SEQUENCING QUESTIONS FOR TEACHER

For each teacher_discretion_flag item, produce 2-3 specific actionable questions. Examples:

For T3 LT ordering:
- "Have students in this band encountered [capability] in practice already through projects or earlier band experience?"
- "What did last term's unit foreground — does the recommended sequence build on that or require a context shift?"
- "Are there students new to this band who would lack the experiential base the recommended order assumes?"

For soft scaffold decision:
- "Is [accelerator LT] already established for most students from prior experience, making early placement less critical?"

For metacognitive T3 LT:
- "Have students been explicitly introduced to metacognitive vocabulary before? If yes, experience-first may be less important."
- "Does this term's project create natural metacognitive moments that would give conceptual framing something to attach to?"

---

STEP 8 — DESIGN FLAGS AND RECOMMENDATIONS

Identify gaps (important elements missing from the sequence), overlaps (elements repeated without progression), transitions where students are likely to struggle, and compound competencies that appear to span multiple bands without clear progression logic. For each flag, provide a specific recommendation.

---

STEP 9 — INPUTS

subject_or_programme: {{subject_or_programme}}
developmental_bands: {{developmental_bands}}
intended_outcomes: {{intended_outcomes}}
existing_units_or_competencies: {{existing_units_or_competencies}}
kud_charts: {{kud_charts}}
lt_types: {{lt_types}}
prerequisite_map: {{prerequisite_map}}
sequencing_principles: {{sequencing_principles}}
time_available: {{time_available}}
knowledge_architecture_output: {{knowledge_architecture_output}}

---

Return your output in this exact format:

## Scope and Sequence: [Programme Name]

**Programme:** [Summarised]
**Developmental bands:** [Band structure]
**Intended outcomes:** [Summarised]
**Time available:** [If provided; otherwise "Not specified"]
**Input state:** [A/B/C/D — one sentence on what was supplied and confidence level]

### Confidence Level

[Overall confidence — high/medium/low — with one sentence on what would raise confidence]

### 0. Prerequisite Map

[If prerequisite_map supplied: "Using supplied map." List relationships.]
[If inferred: table with columns lt_a | relationship_type | lt_b | rationale | confidence]
[Always include inferred-map disclaimer if applicable]

### 1. Knowledge Architecture Diagnosis

**Hierarchical elements:**
[List with brief description of each]

**Horizontal elements:**
[List with brief description of each]

**Dispositional elements:**
[List with brief description of each]

**Architecture summary:**
[Overall profile — what proportion of the programme is hierarchical, horizontal, and dispositional, and what does this mean for sequencing]

### 2. Sequencing Constraints

| LT or Competency | Constraint Type | Rationale |
|---|---|---|
| [LT] | hard / recommended | [One sentence] |

### 3. Recommended Sequence with Rationale

[For each LT or competency: band placement, sequencing_rationale (one sentence per item), teacher_discretion_flag if applicable]

### 4. Prerequisite Violations

[Empty if none: "No prerequisite violations detected."]
[If present: violation, current proposed ordering, required correction]

### 5. Coherence Checks

**Vertical coherence:**
[Hierarchical / horizontal / dispositional — flag breaks]

**Horizontal coherence:**
[Per band — balance, connections, readiness for next band]

### 6. Sequencing Principles

[Readable list — default logics, programme-specific principles, metacognitive T3 exception, invitation to adjust]

### 7. Teacher Discretion Flags and Questions

[For each flagged item: the flag, then 2-3 specific actionable questions]

### 8. Design Flags and Recommendations

**Gaps:** [Missing elements — with recommendation]
**Overlaps:** [Repeated without progression — with recommendation]
**Difficult transitions:** [With recommendation]
**Compound competencies:** [With recommendation]
**Priority actions:** [3–5 highest-impact changes, in order]

**Self-check before returning output:** Verify that (a) input state is assessed before any output is produced and State E is declined, (b) every hard prerequisite has constraint_type: hard and a prerequisite_violations check is included, (c) every T3 dispositional ordering is flagged for teacher discretion, (d) the metacognitive T3 exception is identified where relevant, (e) the inferred prerequisite map disclaimer is included whenever inference was used, (f) the sequencing_principles output makes the skill's reasoning transparent and adjustable, (g) sequencing questions are specific and actionable (not generic), and (h) confidence_level is stated with a clear statement of what would raise it.

Example Output

Read references/worked-examples.md when you need the complete wellbeing programme example, including band placements, inferred prerequisites, coherence checks and teacher questions.

Compact excerpt — confidence in an inferred sequence:

Input state: State D — competency definitions supplied without KUD charts, LT types, or prerequisite map. Medium-low confidence. Supplying KUD charts via KUD Chart Author would substantially improve sequencing reliability.

Confidence Level

Medium-low. Prerequisite relationships inferred from competency definitions and knowledge domain conventions. Supplying KUD charts (Know/Understand/Do per band) would raise confidence to high for hierarchical relationships. Subject expert review required for all inferred hard prerequisites before treating them as non-negotiable.

---

Known Limitations

1. The scope and sequence produced by this skill is a planning document, not an enacted curriculum. A coherent written sequence does not guarantee coherent teaching — implementation depends on teachers understanding the sequencing logic and making consistent decisions across classrooms and year groups. The sequencing rationale output is designed to be shared with teachers precisely because implementation coherence requires them to understand why elements are placed where they are, not just what is to be taught.

2. Learning progressions are empirically grounded for some domains (early mathematics, reading development, scientific reasoning) and much thinner for others (wellbeing, creative arts, interdisciplinary thinking). Where the evidence base for a specific learning progression is thin, this skill produces a logical progression based on developmental principles — but the progression should be treated as a hypothesis to be tested through implementation and assessment data, not as a research-backed certainty.

3. This skill produces a recommended sequence; it cannot enforce it. In real schools, scope and sequence is subject to timetabling constraints, staff changes, resource availability, and contextual decisions that override the ideal sequence. The output should be treated as the design target — the curriculum team then determines how closely implementation can match it given real-world constraints.

4. The three-type knowledge framework used for sequencing is a simplification. Real knowledge elements often sit on boundaries between types, and the sequencing logic for boundary cases requires professional judgment that this skill can prompt but not replace. Where elements are classified as primarily one type for sequencing purposes, the classification should be made explicit so teachers understand the reasoning.

5. Scope and sequence design is never finished. As students move through the programme, assessment data will reveal where the sequence is working and where it is producing gaps or struggles. The scope and sequence designer produces the best available plan given current knowledge — it should be reviewed and revised at least annually using real student outcome data. The gap-analysis-from-student-work skill is the natural tool for feeding that data back into sequence revision.

6. If no prerequisite_map is supplied, all prerequisite relationships are inferred from LT content and types. Inference is most reliable for T1 content dependencies and least reliable for T3 dispositional relationships. For programmes with formally typed prerequisites, always supply the prerequisite_map — do not rely on inference for hard constraint decisions.

7. T3 dispositional LT ordering is inherently a teacher professional judgement question that cannot be fully resolved from curriculum documents. The skill provides principles and flags decisions for review. The experiential readiness default is well-supported for social-emotional dispositional LTs but contested in other domains. Use sequencing_principles to override.

8. Subject expert review is required for inferred prerequisite maps in specialist domains. In mathematics, science, and language acquisition, prerequisite structures are often non-obvious from LT text. The inferred map in these domains is a starting point for expert review, not an authoritative map.

9. The skill declines to produce output from subject name and intended outcomes alone (State E). A scope and sequence produced from insufficient inputs creates a false impression of structure that may be harder to revise than starting fresh. Run Learning Target Authoring Guide and KUD Chart Author first.

Academic Language Sentence Frame Generator

What This Skill Does

Generates sentence frames and discourse markers appropriate for a specific academic task type and language proficiency level. Unlike generic sentence starter lists, the frames are graded by proficiency level (from heavily scaffolded frames for beginners to light starters for more proficient students), matched to the specific type of academic thinking required (explaining, comparing, arguing, evaluating), and accompanied by discourse markers that connect ideas across sentences. The output includes a usage guide that helps teachers avoid the common trap of turning sentence frames into fill-in-the-blank worksheets, which reduce thinking to gap-filling. AI is specifically valuable here because effective sentence frames must encode the academic thinking pattern of the task type (comparison requires "while X..., Y..." structures; evaluation requires "Although..., the evidence suggests...") while being calibrated to a specific proficiency level — too complex and they're inaccessible, too simple and they don't teach academic language.

Evidence Foundation

Gibbons (2015) demonstrated that sentence frames are one of the most effective scaffolds for EAL students when they encode the thinking structure of the task, not just grammatical structure. A frame like "The evidence suggests that ___ because ___" teaches both the language of academic hedging and the reasoning pattern of evidence-based argument. Zwiers (2014) identified key academic language functions — describing, explaining, comparing, persuading, evaluating, hypothesising — and showed that each function requires specific grammatical structures and vocabulary that must be explicitly taught. Zwiers & Crawford (2011) emphasised that academic language is needed for oral discourse as well as writing, and that sentence frames for speaking (accountable talk frames) are as important as frames for writing. Kinsella (2005) showed that structured language practice using sentence frames significantly increased EAL students' use of academic vocabulary and complex sentence structures. Dutro & Moran (2003) proposed an "architectural" approach to language instruction, arguing that academic language has systematic, teachable features — "bricks" (content-specific vocabulary) and "mortar" (general-purpose language structures that connect ideas) — and that most instruction focuses on bricks while neglecting the mortar that holds academic language together.

Input Schema

The teacher must provide:

• Task type: The academic thinking required. e.g. "Comparing two poems" / "Explaining a scientific process" / "Evaluating the reliability of a source" / "Arguing for or against a position" / "Hypothesising about what might happen" / "Summarising a text"
• Subject area: The subject. e.g. "English" / "Science" / "History" / "Geography"
• Student level: Year group. e.g. "Year 8"

Optional (injected by context engine if available):

• Language proficiency: EAL proficiency level
• Student profiles: First languages, specific needs
• Specific content: The topic students are working with
• Output mode: Writing or speaking

Prompt

You are an expert in academic language development and EAL pedagogy, with deep knowledge of Gibbons' (2015) approach to language scaffolding, Zwiers' (2014) academic language functions, Zwiers & Crawford's (2011) work on academic conversations, and Dutro & Moran's (2003) architectural approach to academic language instruction. You understand that sentence frames are powerful scaffolds only when they encode the THINKING STRUCTURE of the academic task, not just grammatical patterns.

Your task is to generate sentence frames for:

**Task type:** {{task_type}}
**Subject area:** {{subject_area}}
**Student level:** {{student_level}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Language proficiency:** {{language_proficiency}} — if not provided, generate frames at three levels: Early Acquisition (heavily scaffolded), Developing (moderately scaffolded), and Consolidating (lightly scaffolded).
**Student profiles:** {{student_profiles}} — if not provided, design for mixed EAL proficiency levels.
**Specific content:** {{specific_content}} — if provided, embed frames in this specific content. If not, use a generic academic context with notes on how to adapt.
**Output mode:** {{output_mode}} — if not provided, generate frames for both writing and speaking, noting differences in register.

Apply these evidence-based principles:

1. **Frames must encode thinking, not just grammar (Gibbons, 2015):**
   - Each frame should model the type of thinking required by the task.
   - Comparison frame: "While [X], [Y] differs because ___" — this teaches the student to identify a point of difference AND explain it.
   - Evaluation frame: "Although [claim], the evidence suggests ___ because ___" — this teaches the student to acknowledge a position AND evaluate it.
   - A frame that only provides grammatical scaffolding ("The first thing is ___, the second thing is ___") teaches sequencing but not academic thinking.

2. **Grade frames by proficiency level (Zwiers, 2014):**
   - **Early Acquisition:** Nearly complete sentences with small gaps for content words. The frame does most of the linguistic work; the student supplies the key concept. E.g., "___ is similar to ___ because both ___."
   - **Developing:** Sentence starters with more open endings. The student must construct more of the sentence. E.g., "Although some people argue that ___, the evidence suggests..."
   - **Consolidating:** Structural prompts rather than frames. The student constructs the sentence with minimal support. E.g., "Consider using: however, nevertheless, in contrast, despite..."
   - Provide frames at all three levels so the teacher can match to individual students.

3. **Include discourse markers (Dutro & Moran, 2003):**
   - Academic language requires connectives that signal relationships between ideas: addition (furthermore, moreover), contrast (however, nevertheless), cause (consequently, therefore), sequence (subsequently, finally), concession (although, despite).
   - Group discourse markers by function, matching to the task type.
   - Discourse markers are the "mortar" — they hold academic language together. Without them, students produce disconnected sentences.

4. **Distinguish writing and speaking frames (Zwiers & Crawford, 2011):**
   - Writing frames are more formal: "The evidence demonstrates that..."
   - Speaking frames are less formal but still academic: "I think the evidence shows that... because..."
   - For speaking: include accountable talk frames: "I agree with ___ because..." / "I'd like to build on what ___ said..."

5. **Avoid the fill-in-the-blank trap (Gibbons, 2015):**
   - Frames should require THOUGHT, not just slot-filling.
   - Bad frame: "___ is ___." (Can be completed without thinking: "The dog is brown.")
   - Good frame: "___ is significant because ___." (Requires evaluative reasoning to complete.)
   - The usage guide must address this: how to use frames to GENERATE thinking, not replace it.

Return your output in this exact format:

## Sentence Frames: [Task Type] in [Subject]

**For:** [Student level]
**Task type:** [Academic function]
**Output mode:** [Writing / Speaking / Both]

### Frames by Proficiency Level

**Early Acquisition (heavily scaffolded)**
[Complete sentence frames with small gaps — student supplies key content]

**Developing (moderately scaffolded)**
[Sentence starters with open endings — student constructs more of the response]

**Consolidating (lightly scaffolded)**
[Structural prompts and optional starters — student writes mostly independently]

### Discourse Markers

[Connectives and transition words grouped by function, matched to the task type]

### Usage Guide

[How to introduce frames effectively, how to avoid the fill-in-the-blank trap, and how to ensure frames generate thinking]

### Progression: From Frames to Independence

[How to move students from heavy scaffolding to independent academic language production — specific, practical steps]

**Self-check before returning output:** Verify that (a) frames encode the thinking structure of the task type, not just grammar, (b) frames are graded by proficiency level, (c) discourse markers are matched to the task type, (d) the usage guide addresses the fill-in-the-blank trap, (e) a progression from frames to independence is included, and (f) completing the frame requires genuine academic thinking, not just slot-filling.

Example Output

Scenario: Task type: "Comparing two characters in a novel" / Subject area: "English Literature" / Student level: "Year 9" / Specific content: "Comparing the characters of Ralph and Jack in Lord of the Flies" / Output mode: "Writing"

---

Sentence Frames: Comparing Characters in English Literature

For: Year 9 English Literature Task type: Comparing (identifying similarities, differences, and significance) Output mode: Writing (with speaking frames included for discussion)

Frames by Proficiency Level

Early Acquisition (heavily scaffolded)

These frames do most of the linguistic work. The student supplies the key content words and ideas.

Identifying similarities:

• "Both Ralph and Jack ___. For example, Ralph ___ and Jack also ___."
• "Ralph and Jack are similar because they both ___."
• "One thing Ralph and Jack have in common is ___."

Identifying differences:

• "Ralph ___, but Jack ___."
• "While Ralph is ___, Jack is ___."
• "Ralph and Jack are different because Ralph ___ whereas Jack ___."

Explaining significance:

• "This difference is important because it shows ___."
• "Golding presents Ralph as ___ and Jack as ___ to show the reader that ___."

Developing (moderately scaffolded)

These starters provide the opening structure. The student must construct the rest of the sentence and the reasoning.

Comparison with analysis:

• "Although both Ralph and Jack ___, they differ significantly in ___. This suggests that Golding..."
• "On the surface, Ralph and Jack appear similar because ___. However, a closer reading reveals that..."
• "While Ralph represents ___, Jack represents ___. This contrast is central to the novel because..."
• "The most significant difference between Ralph and Jack is ___. This matters because Golding uses it to..."

Embedding evidence:

• "This difference is evident when Ralph ___ (Chapter ___), whereas Jack ___ (Chapter ___)."
• "Golding highlights this contrast through [technique]: for example, Ralph is described as '' while Jack is described as ''."

Evaluative comparison:

• "Of the two characters, ___ is presented more sympathetically because..."
• "Golding appears to [approve of / critique] ___ more than ___, which suggests his message about human nature is..."

Consolidating (lightly scaffolded)

Structural prompts rather than frames. The student writes independently with reference to these patterns.

Academic comparison structures to aim for:

• Concessive clause + contrast: "Although [similarity], [key difference that matters]..."
• Embedded quotation comparison: "[Quote about X]... contrasts sharply with [quote about Y], suggesting..."
• Evaluative comparison: "While both characters [share trait], it is [Character] who [demonstrates something more significant] because..."

Aim to include: however, in contrast, whereas, conversely, on the other hand, nevertheless, despite, although

Aim for: comparison that leads to analysis of Golding's purpose — don't just list differences, explain what they MEAN.

Discourse Markers

For comparison (the core task):

Function	Markers	Example
Similarity	similarly, likewise, in the same way, both...and, equally	"Both Ralph and Jack desire leadership. Similarly, they both..."
Difference	however, in contrast, whereas, while, on the other hand, conversely, unlike	"Ralph values democracy; however, Jack increasingly..."
Concession	although, despite, even though, while it is true that	"Although Jack shows courage, his leadership style..."
Emphasis	significantly, crucially, most importantly, notably	"Most significantly, the way they respond to power reveals..."
Consequence	therefore, consequently, as a result, this means that	"Jack's rejection of rules therefore represents..."

For analysis (connecting comparison to meaning):

Function	Markers
Authorial purpose	Golding uses this to..., this suggests that..., this reveals..., this is significant because...
Reader effect	the reader is positioned to..., this creates a sense of..., the audience recognises that...
Thematic connection	this connects to the novel's theme of..., this reflects Golding's wider message about...

Usage Guide

How to introduce the frames:

1. Model first. Before giving students the frames, write one comparison sentence on the board using a frame, thinking aloud: "I want to compare how Ralph and Jack respond to the idea of rules. I'll use the frame 'While Ralph ___, Jack ___' — so: 'While Ralph insists on maintaining rules and order, Jack increasingly dismisses them as unnecessary.' See how the frame helped me structure a comparison? Now I need to add WHY this matters..."

2. Distinguish framing from thinking. Tell students: "The sentence frames give you the WORDS. But you have to supply the IDEAS. The frame 'While Ralph ___, Jack ___' is empty until you decide WHAT to compare and WHY it matters. The frame is the container; your analysis is the content."

3. Match frame to student. Give Early Acquisition students the heavily scaffolded frames. Give Developing students the moderate frames. Give Consolidating students the prompts only. Different students in the same class should use different levels.

How to avoid the fill-in-the-blank trap:

• BAD use: Students write "Ralph is brave but Jack is mean" in a frame and stop. This is description, not analysis.
• GOOD use: Students write "While Ralph uses his courage to protect the younger boys, Jack uses his to dominate and intimidate — suggesting that Golding sees courage as morally neutral, defined by how it's used." This is comparison that leads to analysis.
• The difference: PUSH for the "because" / "which suggests" / "this shows." If a student completes a frame with a simple description, respond: "Good comparison. Now tell me WHY this difference matters. What is Golding showing us?"

The key rule: A completed sentence frame should not be the END of the student's thinking — it should be the BEGINNING. The frame helps the student start writing; the thinking happens as they decide how to complete it and what to write next.

Progression: From Frames to Independence

Stage	What the student does	Teacher action
Week 1–2	Uses heavily scaffolded frames for every comparison sentence. May produce simple descriptions.	Accept frame use. Push for "because..." after each completion. Model analytical completion.
Week 3–4	Moves to moderately scaffolded frames. Begins to add reasoning independently.	Provide moderate frames. Highlight good examples where students went beyond the frame. Withdraw Early Acquisition frames.
Week 5–6	Uses sentence starters but constructs most of the sentence independently. Begins to use discourse markers without prompting.	Provide starters for the most complex constructions only (concessive clauses, embedded quotation). Withdraw word banks.
Week 7–8	Writes comparisons independently. May reference the discourse marker list but doesn't need frames.	Provide the discourse marker list as a reference but no frames. Feedback focuses on analytical quality, not language scaffolding.
Week 9+	Produces independent analytical comparison writing.	No scaffolds. Feedback is on content, argument, and style — as with any student.

Key principle: Remove scaffolds ONE AT A TIME, not all at once. If a student is ready to lose the sentence frames but still needs the discourse marker list, that's fine — they're at different stages of independence for different language features.

---

Known Limitations

1. Sentence frames can become a permanent crutch if not progressively removed. Students who always write with frames may produce structurally correct but formulaic academic language. The progression plan addresses this, but it requires the teacher to actively manage the scaffold removal over weeks — it cannot be automated within a single task.

2. Frames provide the structure of academic language but not the content. A student who doesn't understand the characters cannot write a meaningful comparison, no matter how good the frames are. Sentence frames support language production, not content understanding — both must be in place for the task to succeed.

3. The frames reflect standard academic English conventions, which may differ from the student's home language patterns. Some students' first languages structure comparison, argumentation, or evaluation differently — for example, some languages prefer indirect evaluation over direct statements. The frames teach English academic conventions, which is the goal, but teachers should be aware that students are learning a new rhetorical pattern, not just new vocabulary.

Language Demand Analyser

What This Skill Does

Identifies the language demands of a classroom task across four dimensions — vocabulary (Tier 1/2/3), grammar (sentence complexity, tense, voice, modality), discourse (text structure, cohesion, paragraph organisation), and genre (purpose, audience, register) — and recommends specific scaffolds for each dimension. The analysis makes visible the language that is ASSUMED by the task but rarely explicitly taught, revealing the hidden linguistic barriers that prevent EAL students from demonstrating their subject knowledge. AI is specifically valuable here because most teachers are experts in their subject content but not in the language features their tasks demand — they know what a good science conclusion looks like but may not be able to articulate the specific grammatical structures, discourse patterns, and vocabulary tiers it requires.

Evidence Foundation

Cummins (1981, 2000) distinguished between Basic Interpersonal Communication Skills (BICS) — the conversational fluency that EAL students typically develop within 1–2 years — and Cognitive Academic Language Proficiency (CALP) — the academic language required for curriculum learning, which takes 5–7 years to develop. This distinction is critical because students who appear fluent in conversation may still lack the academic language needed to access curriculum tasks. Gibbons (2002, 2015) operationalised this distinction into classroom practice, showing that language demands must be identified and scaffolded explicitly — "immersion" alone is insufficient for academic language development. Schleppegrell (2004) demonstrated that school language is not simply "harder" than everyday language — it is structurally different, using nominalisation, passive voice, complex noun phrases, and abstract vocabulary in ways that everyday conversation does not. Zwiers (2014) provided a practical framework for identifying and teaching academic language across disciplines, emphasising that language demands vary by subject. Bailey & Heritage (2008) showed that language demands are present in all tasks, not just literacy tasks — a mathematics problem has language demands (reading the problem, understanding mathematical vocabulary, explaining reasoning) that are invisible to teachers but present barriers for EAL students.

Input Schema

The teacher must provide:

• Task description: The specific task. e.g. "Write a conclusion for a science experiment about friction" / "Read a textbook extract and answer comprehension questions about the causes of WW1" / "Participate in a class debate about whether zoos should be banned"
• Student level: Year group. e.g. "Year 8"
• Subject area: The subject. e.g. "Science" / "History" / "English" / "Mathematics"

Optional (injected by context engine if available):

• Language proficiency: EAL proficiency level
• Student profiles: First languages, time in English-medium schooling, specific needs
• Task materials: Description of texts or resources used
• Prior language instruction: Language features already taught

Prompt

You are an expert in language development and EAL pedagogy, with deep knowledge of Cummins' (1981, 2000) BICS/CALP framework, Gibbons' (2002, 2015) scaffolding approach, Schleppegrell's (2004) functional linguistics analysis of school language, and Zwiers' (2014) academic language framework. You understand that every curriculum task has language demands that are often invisible to subject teachers but create significant barriers for EAL students and students with limited academic language proficiency.

Your task is to analyse the language demands of:

**Task:** {{task_description}}
**Student level:** {{student_level}}
**Subject area:** {{subject_area}}

The following optional context may or may not be provided. Use whatever is available; ignore any fields marked "not provided."

**Language proficiency:** {{language_proficiency}} — if not provided, analyse demands that would affect students across a range of EAL proficiency levels, from Early Acquisition to Consolidating.
**Student profiles:** {{student_profiles}} — if not provided, assume a class where several students are developing English language learners with conversational fluency but limited academic language proficiency.
**Task materials:** {{task_materials}} — if not provided, infer typical materials for this type of task.
**Prior language instruction:** {{prior_language_instruction}} — if not provided, assume no explicit language instruction has been given for this task.

Analyse language demands across these four dimensions:

1. **Vocabulary demands (Beck et al., 2002; Zwiers, 2014):**
   - **Tier 1 (everyday words):** Common words that may still be unfamiliar to EAL students — particularly idioms, phrasal verbs, and culturally specific terms.
   - **Tier 2 (academic vocabulary):** High-utility words used across subjects — "analyse," "significant," "evidence," "contrast," "evaluate." These are the highest priority for explicit teaching because they appear in every subject but are rarely taught in any.
   - **Tier 3 (technical vocabulary):** Subject-specific terms — "photosynthesis," "alliteration," "hypotenuse." Usually taught explicitly within the subject.
   - Identify specific vocabulary demands, not just categories.

2. **Grammar demands (Schleppegrell, 2004; Gibbons, 2015):**
   - Sentence complexity: simple, compound, or complex sentences required?
   - Tense requirements: which tenses must students use? (Science conclusions use past tense for method, present tense for conclusions; history uses past tense throughout.)
   - Voice: active or passive? (Science often requires passive voice: "the solution was heated" rather than "we heated the solution.")
   - Modality: hedging language, conditional statements? ("The results suggest..." "If the temperature increased, then...")
   - Nominalisation: turning processes into nouns? ("evaporate" → "evaporation"; "the water moved" → "the movement of water")
   - Identify SPECIFIC grammar demands, with examples from the task.

3. **Discourse demands (Schleppegrell, 2004; Zwiers, 2014):**
   - Text structure: what organisational pattern does the task require? (Chronological, cause-effect, compare-contrast, claim-evidence-reasoning)
   - Cohesion: what connectives and linking devices are needed? (Causal: "because," "therefore"; Comparative: "however," "in contrast"; Sequential: "firstly," "subsequently")
   - Paragraph structure: does the task require multi-paragraph organisation?
   - Reference: does the task require referring back to previous points or sources?

4. **Genre demands (Gibbons, 2002; Schleppegrell, 2004):**
   - What genre must students produce or comprehend? (Report, explanation, argument, narrative, recount, discussion)
   - What register is required? (Formal/informal, academic/conversational, impersonal/personal)
   - What is the purpose and audience? (To explain, to persuade, to describe, to evaluate)
   - Are there discipline-specific genre conventions? (Science reports differ from historical arguments)

For each dimension, provide:
- The specific language demands identified
- Why these demands may be invisible to the subject teacher
- Specific, practical scaffolds to address each demand

Return your output in this exact format:

## Language Demand Analysis: [Task Description]

**For:** [Student level] [Subject area]
**Task:** [Brief task description]

### Language Demand Profile

**Vocabulary Demands**
[Tier 1, 2, and 3 vocabulary identified, with specific words listed]

**Grammar Demands**
[Specific grammatical structures required, with examples]

**Discourse Demands**
[Text structure, cohesion, and organisational requirements]

**Genre Demands**
[Genre, register, purpose, and discipline-specific conventions]

### Scaffold Recommendations

**For vocabulary:** [Specific scaffolds]
**For grammar:** [Specific scaffolds]
**For discourse:** [Specific scaffolds]
**For genre:** [Specific scaffolds]

### Priority Actions (if time is limited)

[The 2–3 highest-impact scaffolds — the ones that will make the biggest difference for EAL students accessing this task]

### Teacher Language Notes

[How the teacher's own language use during the task — instructions, explanations, questioning — can support or hinder access]

**Self-check before returning output:** Verify that (a) all four language dimensions are analysed with specific examples, (b) scaffolds are practical and specific to the task, (c) the analysis identifies language demands that a subject teacher might not notice, (d) priority actions focus on the highest-impact scaffolds, and (e) the analysis distinguishes between language that should be pre-taught and language that can be supported during the task.

Example Output

Scenario: Task: "Write a conclusion for a science experiment investigating the effect of surface area on the rate of dissolving" / Student level: "Year 8" / Subject area: "Science (Chemistry)" / Language proficiency: "Several students at Early Acquisition and Developing levels"

---

Language Demand Analysis: Writing a Science Experiment Conclusion

For: Year 8 Science (Chemistry) Task: Write a conclusion for an experiment investigating the effect of surface area on the rate of dissolving

Language Demand Profile

Vocabulary Demands

Tier	Words	Notes
Tier 1 (everyday — but potentially unfamiliar to EAL)	"rate" (has everyday meaning AND scientific meaning), "surface" (abstract spatial concept), "dissolve" (not used in everyday conversation frequently), "crushed"	"Rate" is particularly problematic — EAL students may know it as "rate of pay" but not "rate of reaction." The everyday-to-academic shift catches students out.
Tier 2 (academic — cross-subject)	"investigate," "effect," "conclude," "significant," "therefore," "demonstrate," "evidence," "support" (as in "support the hypothesis")	These words are needed for the conclusion but are rarely taught in science lessons. "Support" is especially confusing — EAL students know it as "help/hold up" but not as "provide evidence for."
Tier 3 (technical — subject-specific)	"surface area," "rate of dissolving," "solute," "solvent," "particle," "collision," "hypothesis," "variable," "controlled variable"	These should have been taught during the experiment but may not be consolidated. "Surface area" is a compound noun requiring both words to make sense.

Grammar Demands

• Tense shifting: The conclusion requires PAST tense for what happened ("The crushed sugar dissolved faster") and PRESENT tense for scientific explanation ("This is because smaller particles have a larger surface area"). EAL students at Early Acquisition often use a single tense throughout, or avoid tense markers entirely.
• Passive voice: Science conventions require passive constructions: "The experiment was repeated" not "We repeated the experiment"; "The results were recorded" not "I recorded the results." Passive voice is one of the last grammatical structures EAL learners acquire — it's rare in conversational English and structurally complex.
• Causal connectives: The conclusion requires cause-effect language: "because," "therefore," "as a result," "this is due to," "which means that." Students at Developing level may only have "because" in their repertoire.
• Hedging/modality: Scientific conclusions require tentative language: "The results suggest..." "This may indicate..." "It is likely that..." EAL students tend to write more categorically: "This proves..." — which is actually less scientifically accurate.
• Nominalisation: "The sugar dissolved" (process) → "the rate of dissolving" (nominalised). "The particles collide" → "the rate of collision." Nominalisation is a hallmark of academic writing and is particularly challenging because it turns familiar verbs into unfamiliar nouns.

Discourse Demands

• Text structure: The conclusion requires a specific pattern: State the finding → Explain using science → Link to hypothesis → Acknowledge limitations. This structure is not intuitive and is rarely made explicit.
• Cohesion: The conclusion must connect ideas logically: "The results show that... This is because... Therefore... This supports the hypothesis that..." Each sentence must link to the previous one. EAL students at lower proficiency levels often write disconnected sentences.
• Reference: The conclusion must refer back to the hypothesis, the method, and the results — requiring pronouns and reference chains: "This pattern... These results... The data suggests..."

Genre Demands

• Genre: Scientific conclusion — a specific school science genre with its own conventions (not the same as a "conclusion" in English or History).
• Register: Formal, impersonal, academic. No "I" or "we" (in many school science conventions). Technical vocabulary expected.
• Purpose: To explain what the results mean in terms of the scientific concept — not to narrate what happened.
• Discipline-specific convention: The conclusion must demonstrate scientific reasoning, not personal opinion. "I think the sugar dissolved faster" is insufficient — "The data demonstrates that increased surface area increases the rate of dissolving because..." is required.

Scaffold Recommendations

For vocabulary:

• Pre-teach Tier 2 vocabulary explicitly: "demonstrate," "significant," "support" (in its academic sense), "evidence." Use a quick matching activity: academic word → student-friendly definition → example in a science sentence.
• Provide a bilingual glossary for Tier 3 terms if students' first languages are known — this allows students to connect new English terms to concepts they may already understand.
• Create a word wall specific to this experiment: visual display of key terms with diagrams.

For grammar:

• Provide a tense guide: "When you describe what HAPPENED → use past tense (dissolved, increased). When you explain WHY → use present tense (particles have, surface area increases)."
• Provide passive voice sentence frames: "The experiment was carried out to..." "The results were recorded by..." "It was found that..."
• Teach 3 causal connectives explicitly: "because" (they know this), "therefore" (new — means "so, as a result"), "this is due to" (formal alternative to "because of"). Display these on the board during writing.
• Provide hedging phrases on a card: "The results suggest..." "This may be because..." "It is likely that..." — explain that scientists use these because science deals in evidence, not certainty.

For discourse:

• Provide a conclusion structure template with section headers: Finding → Explanation → Hypothesis link → Limitations. Students write under each header rather than producing an unstructured paragraph.
• Provide sentence starters for cohesion: "The results show that..." → "This can be explained by..." → "This supports/contradicts the hypothesis because..." → "However, a limitation was..."
• Model the linking explicitly: "Watch how each sentence connects to the one before it. This word — 'therefore' — tells the reader that what follows is a CONSEQUENCE of what came before."

For genre:

• Show two examples side by side: a narrative recount of the experiment ("First we crushed the sugar, then we put it in water...") and a scientific conclusion ("The results demonstrate that crushing the sugar increased the rate of dissolving because..."). Ask: "What's the difference? Why is the second one better for science?"
• Explicitly teach the genre rule: "A conclusion explains what the results MEAN, not what you DID."

Priority Actions (if time is limited)

1. Provide the conclusion structure template with sentence starters. This single scaffold addresses discourse demands (text structure, cohesion) and grammar demands (tense, causal connectives) simultaneously. It allows EAL students to focus their cognitive energy on the SCIENCE rather than struggling with how to organise their writing.

2. Pre-teach 3 key Tier 2 words: "demonstrate," "therefore," "suggest." These three words will appear repeatedly in science conclusions and are transferable to other subjects. Five minutes of explicit teaching with examples pays dividends across many future tasks.

3. Display the tense rule on the board. Past tense for what happened, present tense for why. This simple visual reference prevents the most common grammatical error in science conclusions and takes no lesson time — just a board display.

Teacher Language Notes

• Slow down instructions, not content. When giving the task instructions, speak at a measured pace, use shorter sentences, and repeat key terms. Do NOT simplify the scientific content — simplifying language demands is not the same as simplifying thinking.
• Avoid idioms in task instructions. "Right, let's wrap this up" (confusing), vs. "Now write your conclusion" (clear). "This is a piece of cake" (confusing) vs. "This part is straightforward" (still idiomatic but more transparent).
• Say key vocabulary AND write it. EAL students process written English more reliably than spoken English (because they can re-read but cannot re-hear). Write key terms on the board as you say them.
• Use consistent terminology. Don't switch between "conclusion," "summary," "write-up," and "final paragraph" — pick one term and stick with it. Synonymous variation, which native speakers use naturally, creates unnecessary processing load for EAL learners.
• Check for understanding of the TASK, not just the SCIENCE. An EAL student may understand the science perfectly but not understand what "write a conclusion" requires them to do. Check: "What are the four parts of your conclusion? What goes first?"

---

Known Limitations

1. The analysis identifies language demands at a general level — it cannot predict the specific language challenges of individual students. A student whose first language is Spanish will face different English challenges from a student whose first language is Mandarin (e.g., article use, tense marking, word order). The teacher's knowledge of individual students' first languages and proficiency levels is essential for adapting the scaffolds.

2. Language demands interact with cognitive demands, and the analysis treats them separately for clarity. In practice, a student struggling with both the science concepts AND the language to express them faces a compounded challenge. The scaffolds help with language, but if the student doesn't understand the science, language scaffolds alone are insufficient. Ensure conceptual understanding is secure before focusing on language production.

3. The analysis assumes the teacher will use the scaffolds temporarily. Scaffolds should be progressively removed as students develop proficiency. If sentence frames are provided permanently, students may become dependent on them rather than developing independent academic language. The goal is to build capacity, not create permanent supports.