Capture saved assessment evidence locally
Change
Saved assessments now offer a local evidence selection: choose up to eight manifest documents and capture their exact six-field references with the returned snapshot ID. Editable saved-ID input and the currently selected release request cannot replace the captured origin. The selection is explicitly unsaved, unvalidated with the server, and not sent to chat; existing API requests remain unchanged.
Capture requires explicit non-superseded saved evidence and complete valid lineage. A new assessment operation revokes eligibility synchronously before token acquisition and clears selection; scope leave/return, delayed responses, unavailable evidence, and replacement snapshots cannot restore it. API-owned schemas are source-pinned to recoupable/api@1de332ebeae6b74c6f21866482175d33cd843009 with only documented import-path adaptations, rather than independently defined wire types.
Validation
- 39 focused unit/mounted tests passed; final five selection tests rerun after a singular/plural copy correction.
- Next route type generation, full TypeScript, scoped ESLint, formatting and diff checks passed.
- Independent correctness and standards reviews: no actionable findings; vendor hashes independently verified.
- Local browser harness mounted the actual CompanyAssessment and request hook with synthetic responses: keyboard selection, eight-document limit, 44px controls, exact capture, pending invalidation, late response after scope leave/return, historical unknown freshness, and 390px wrapping passed. No horizontal overflow at 390px.
The final 320px reload stalled in the browser runner, so that viewport is unverified. Native screenshots predate only the final singular/plural status correction, which has mounted-test coverage. This PR does not enable selected-evidence API validation, production calls, model sends, persistence, or server authorization. API #1208 owns backend validation; root owns release sequencing.
Refs #2224.
Summary by cubic
Adds a local evidence selection to saved assessments: choose up to eight manifest documents and capture their exact six-field references with the snapshot ID. The selection is explicitly unsaved, unvalidated with the server, and not sent to chat; existing API requests are unchanged.
Capture requires a non-superseded saved snapshot with complete valid lineage, and the editable saved-ID input or current release request never replaces the captured origin. Any assessment operation revokes capture eligibility synchronously before token acquisition, so scope changes, delayed responses, and replacement snapshots can't restore it. API-owned schema sources are vendored and pinned in source-pin.json with only import-path adaptations.
Summary by CodeRabbit
- New Features
- Eligible saved company assessments now offer a way to select up to eight evidence documents and capture the selection for review as read-only JSON.
- You can clear a selection; it becomes unavailable if the assessment is busy, outdated, or otherwise ineligible.
- Evidence selections are local and are not saved or sent to chat.