
Investigating Anomalous Results
Structured protocol for investigating surprising, impossible, or failing results before any data, code, or parameter gets adjusted.
- GitHub stars
- 202
- last updated
- Jun 1, 2026
What it does
- Covers four phases: characterize the anomaly, compare against a working reference, test one hypothesis, and resolve by cause (code bug, data issue, or real finding)
- Enforces a rule that no data drops, parameter changes, or re-runs happen until the root cause is identified and documented
- Includes red flags, common rationalizations, and stakeholder warning signs that signal a result is being tweaked rather than explained
- Designed for analysts and researchers who face convenient or suspicious results and need a defensible path to resolution
Install
$ npx -y skills add k-dense-ai/science-superpowers --skill investigating-anomalous-results -yInstalls into .agents/skills for every agent it finds here, and .claude/skills for Claude Code.
Files
Investigating Anomalous Results
Overview
Random tweaks waste time and manufacture false findings. Quietly dropping the inconvenient data point, nudging the cutoff, or re-running until it "works" doesn't fix the problem — it fabricates a result.
Core principle: ALWAYS find the root cause before adjusting anything. An adjustment made before you understand the cause is, at best, noise and, at worst, fraud.
Violating the letter of this process is violating the spirit of it.
The first question for any anomaly: is this a code bug, a data issue, or a real finding? You cannot answer by guessing, and you must not "fix" it until you know — because a real finding is not something to fix.
The Iron Law
NO ADJUSTMENTS WITHOUT ROOT-CAUSE INVESTIGATION FIRST
No dropping data, changing a test, transforming a variable, re-running with new parameters, or "cleaning" until Phase 1 is complete. If you haven't found the cause, you cannot justify the adjustment.
When to Use
Details
- Repository
- github.com/k-dense-ai/science-superpowers
- First seen
- Jun 11, 2026
More from k-dense-ai/science-superpowers
All 15- Preregistering Analysis
- Receiving Critical Review
- Reporting And Archiving Findings
- Requesting Red Team Review
- Setting Up Reproducible Analysis
- Subagent Driven Analysis
- Surveying Prior Work
- Using Science Superpowers
- Verifying Results Before Claiming
- Writing Science Skills
- Designing The Analysis
- Dispatching Parallel Investigations
- Executing Analysis
- Framing Research Questions
Similar to Investigating Anomalous Results
cavemanjuliusbrussee/caveman630.7K installsUltra-compressed communication mode. Cuts output tokens 65% (measured) by speaking like caveman while keeping full technical accuracy. Supports intensity levels
grill-memattpocock/skills1.6M installsA relentless interview to sharpen a plan or design.
improveshadcn/improve42.5K installsSurvey any codebase as a senior advisor and produce prioritized, self-contained implementation plans for OTHER models/agents to execute. Strictly read-only on s
systematic-debuggingobra/superpowers336.3K installsUse when encountering any bug, test failure, or unexpected behavior, before proposing fixes
karpathy-guidelinesforrestchang/andrej-karpathy-skills17.9K installsBehavioral guidelines to reduce common LLM coding mistakes through explicit assumptions, simplicity, and verifiable success criteria.
find-skillsvercel-labs/skills4.4M installsHelps users discover and install agent skills when they ask questions like "how do I do X", "find a skill for X", "is there a skill that can...", or express int