This sits between raw detection output and strategy implementation, clustering ticket evidence into reusable edge concepts with explicit thesis statements and invalidation signals. You point it at a directory of ticket YAMLs from your detector run, and it outputs a structured edge_concepts.yaml with support statistics, abstract hypotheses, and export readiness flags. The deduplication logic merges overlapping concepts based on containment thresholds, which helps avoid the ticket-to-strategy overfitting problem when you're dealing with dozens of similar signals. It's concept-level review before you commit to writing actual strategies. The hint promotion feature is clever for bootstrapping when ticket support is thin, though the synthetic ratio cap keeps you honest about what's actually validated.
npx skills add https://github.com/tradermonty/claude-trading-skills --skill edge-concept-synthesizer