Fixed
- Unified research-skill examples, planning output, evaluation artifacts, and runtime evidence submission on the shared claim and checklist-receipt schemas.
- Kept the default MCP decision text to the role-based summary while retaining additional finalists, rejections, reasons, and provenance in structured/widget detail.
- Quarantined unsafe native offers individually so one hostile marketplace listing cannot suppress safe peers from the same store.
- Removed lifecycle reminder activity from merchant-trust evidence and made lossy profile-effect projection visible.
- Retracted the unsupported routine-use host/model claim; no combination is currently qualified.
Removed
- Removed the undiscoverable local Decisions dashboard tab. Audit-backed decision state remains for exact attribution and confirmed outcomes; the MCP host remains the decision surface.