[margrop]

Summary

• add a repository-only scoring-rule schema and a public Score v1 result contract
• score validated attempts across evidence coverage, acquisition order, structured conclusion, counterevidence, and safety
• render total, band, dimension progress, auditable findings, and next-run feedback in the local workbench
• complete P2-004 documentation, TODO, accessibility, and static-output contracts

Safety boundary

• scoring is deterministic and never uses AI or free-text semantic matching
• the client does not import the internal answer or canonical attempt
• all data remains synthetic; scoring has no network, storage, logging, or real-system side effects
• risky restart/write/delete choices only produce local penalties and never execute an action

Validation

• npm run validate
• 15 test files / 178 tests passed
• production Astro build passed
• 38 static contracts passed