What problem does it solve?
Consolidates debate and review artifacts to surface per-pair quality metrics, surface regressions, and provide an evidence-backed view of agent pair performance so teams can prioritize tuning and interventions.
Core Features & Use Cases
- Aggregate debate metadata: Read .ratchet/debates/*/meta.json to compute total debates, consensus rates, rounds, verdict breakdowns, and fast-path usage per pair.
- Ingest review signals: Read .ratchet/reviews/<pair>/review-*.json to compute average effectiveness, gen vs adv splits, and top missed issues.
- Persist and smooth history: Update .ratchet/scores.yaml using an exponential moving average (α=0.3) so pair metrics survive archival and support trend analysis.
- Operational use cases: Track pair health over time, decide when to tighten agents, drill down into recent debates, and guide next debate runs.
Quick Start
Run /ratchet:score to display quality metrics and trends for the current workspace or provide a pair name to focus on a single pair.