stella-ops.org/git.stella-ops.org
1
0
Fork 0
Code Issues Pull Requests Actions 2 Releases Wiki Activity

Repair live AdvisoryAI search corpus parity

Browse Source
This commit is contained in:
master
2026-03-08 13:33:21 +02:00
parent f9d652da32
commit 3e531f0b9e
7 changed files with 299 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

77
docs-archived/implplan/SPRINT_20260308_002_FE_search_live_browser_reproduction_and_fix.md Normal file
View File

@@ -0,0 +1,77 @@
# Sprint 20260308-002 - FE Search Live Browser Reproduction And Fix
## Topic & Scope
- Reproduce the reported global-search failure in a live browser session instead of relying on prior mocked regressions.
- Identify why searches are not returning usable results across pages and fix the concrete operator-facing failure.
- Add focused regression coverage so live search interaction stays usable on primary routes.
- Working directory: `src/Web/StellaOps.Web`.
- Expected evidence: live Playwright/browser reproduction, focused FE tests, and updated UI docs if behavior changes.
## Dependencies & Concurrency
- Depends on the archived search rollout and clarify-guidance sprints.
- Safe parallelism: avoid unrelated Router, trust-admin, offline-kit, and audit-bundle worktree changes.
- Cross-module allowance: documentation updates may touch `docs/modules/ui/**` if the operator search contract changes.
## Documentation Prerequisites
- `src/Web/StellaOps.Web/AGENTS.md`
- `docs/modules/ui/search-self-serve-contract.md`
- `docs/modules/ui/search-zero-learning-primary-entry.md`
## Delivery Tracker
### FE-SEARCH-LIVE-001 - Reproduce the operator failure in a real browser session
Status: DONE
Dependency: none
Owners: Developer (FE), QA
Task description:
- Use Playwright against the live `stella-ops.local` shell to reproduce the reported inability to search from real pages.
- Capture the exact failing interaction path, route context, and UI/network behavior so the fix targets the real defect.
Completion criteria:
- [x] Failure reproduced or disproved on live pages using Playwright.
- [x] Root cause narrowed to a concrete UI, request, or data-readiness issue.
- [x] Execution log records the live reproduction evidence.
### FE-SEARCH-LIVE-002 - Fix the live search interaction failure
Status: DONE
Dependency: FE-SEARCH-LIVE-001
Owners: Developer (FE)
Task description:
- Implement the smallest frontend fix that restores usable search from the live shell.
- Prefer behavior that works without requiring the operator to learn special search syntax or page-specific rituals.
Completion criteria:
- [x] Global search returns usable results or explicit grounded fallback on affected routes.
- [x] The fix does not break prior search guidance/answer behavior.
- [x] Focused tests cover the repaired interaction.
### FE-SEARCH-LIVE-003 - Verify and close
Status: DONE
Dependency: FE-SEARCH-LIVE-002
Owners: QA, Documentation, Project Manager
Task description:
- Re-run browser validation after the fix, add any missing regression coverage, and archive the sprint when complete.
Completion criteria:
- [x] Focused automated coverage passes.
- [x] Execution log records live browser verification after the fix.
- [x] Sprint is archived only after all tasks are DONE.
## Execution Log
| Date (UTC) | Update | Owner |
| --- | --- | --- |
| 2026-03-08 | Sprint created to reproduce and fix the reported live search failure from the operator shell. | Developer |
| 2026-03-08 | Live Playwright reproduction on `https://stella-ops.local` confirmed the operator report: `POST /api/v1/search/query` returned `200` with zero indexed matches on `/releases/versions`, `/security/triage`, `/ops/policy`, and `/ops/operations/doctor`, while `POST /api/v1/search/suggestions/evaluate` returned `404`. Direct inspection of the compose-backed AdvisoryAI service at `127.1.0.44` showed knowledge rebuild output `documentCount=0`, `apiOperationCount=0`, and `doctorProjectionCount=0`, plus stale unified rows with unusable live FTS behavior. A runtime database backfill on `advisoryai.kb_chunk.body_tsv_en` restored exact-term search from the browser for indexed entities such as `critical findings`, `CVE-2024-21626`, `CVE-2025-1001`, and `Deny critical vulnerabilities in production`, but generic Doctor/Release suggestions still fail because the compose service lacks knowledge/API/Doctor corpus parity and does not expose `POST /v1/search/suggestions/evaluate`. | Developer |
| 2026-03-08 | Closure verification completed after the AdvisoryAI runtime parity repair landed in `SPRINT_20260308_003_AdvisoryAI_live_search_compose_parity.md`: live Playwright on `https://stella-ops.local` now returns grounded answers on `/releases/versions`, `/security/triage`, `/ops/policy`, and `/ops/operations/doctor`, and release suggestion chips such as `What blocked this promotion?` execute into grounded answer states. No FE-specific code change was required once the compose-backed AdvisoryAI service shipped the missing docs/OpenAPI/Doctor corpus. | Developer |
## Decisions & Risks
- Decision: validate the report in a live browser session before making any further search UX changes.
- Risk: prior mocked regression coverage may not reflect the real runtime path on `stella-ops.local`.
- Mitigation: reproduce on the live shell first, then add regression coverage for the exact failure mode.
- Risk: the remaining live failure is not primarily in FE scope. The compose-backed AdvisoryAI service behind `stella-ops.local` is running without knowledge/API/Doctor ingestion parity and without the suggestion-viability endpoint that current FE logic expects.
- Mitigation: keep FE fallback work isolated, but track the remaining repair as an AdvisoryAI/runtime follow-up before claiming full live-search recovery.
- Decision: close this sprint without FE code changes because the verified operator failure was caused by compose runtime corpus omission, not by a surviving shell interaction defect.
## Next Checkpoints
- 2026-03-08: live browser reproduction captured.
- 2026-03-08: fix implemented and verified if root cause is in FE scope.

80
docs-archived/implplan/SPRINT_20260308_003_AdvisoryAI_live_search_compose_parity.md Normal file
View File

@@ -0,0 +1,80 @@
# Sprint 20260308-003 - AdvisoryAI Live Search Compose Parity
## Topic & Scope
- Repair the compose-backed AdvisoryAI service behind `https://stella-ops.local` so live search has the same corpus and endpoint surface as the source-run validation lane.
- Restore knowledge/API/Doctor ingestion parity and the suggestion-viability endpoint required by the current search UX.
- Verify the repaired service from the real browser on `stella-ops.local` instead of relying on isolated backend tests.
- Working directory: `src/AdvisoryAI`.
- Expected evidence: focused AdvisoryAI/runtime verification, live browser search checks, and updated docs if setup/runtime behavior changes.
## Dependencies & Concurrency
- Depends on [SPRINT_20260308_002_FE_search_live_browser_reproduction_and_fix.md](/C:/dev/New%20folder/git.stella-ops.org/docs/implplan/SPRINT_20260308_002_FE_search_live_browser_reproduction_and_fix.md) for the live reproduction evidence.
- Safe parallelism: avoid unrelated Router, Web trust-admin, offline-kit, and audit-bundle changes.
- Cross-module allowance: docs under `docs/modules/advisory-ai/**` and `docs/modules/ui/**` may be updated if live-search setup or operator guidance changes; repo-root `.dockerignore` may be updated only as needed to let the AdvisoryAI image package its documented search corpus subset.
## Documentation Prerequisites
- `src/AdvisoryAI/AGENTS.md`
- `src/AdvisoryAI/__Tests/INFRASTRUCTURE.md`
- `docs/modules/advisory-ai/knowledge-search.md`
- `docs/modules/ui/search-zero-learning-primary-entry.md`
## Delivery Tracker
### AI-LIVE-SEARCH-001 - Restore compose AdvisoryAI endpoint parity
Status: DONE
Dependency: none
Owners: Developer (AdvisoryAI)
Task description:
- Determine why the compose-backed AdvisoryAI service at `advisoryai.stella-ops.local` returns `404` for `POST /v1/search/suggestions/evaluate` even though the current source maps that endpoint.
- Repair the runtime so the live service exposes the same search surface as the source-run validation lane.
Completion criteria:
- [x] `POST /v1/search/suggestions/evaluate` returns a valid response from the compose-backed service.
- [x] Root cause is documented in the execution log.
- [x] Any required docs are updated.
### AI-LIVE-SEARCH-002 - Restore compose knowledge corpus parity
Status: DONE
Dependency: AI-LIVE-SEARCH-001
Owners: Developer (AdvisoryAI), QA
Task description:
- Determine why compose rebuilds return `documentCount=0`, `apiOperationCount=0`, and `doctorProjectionCount=0`.
- Repair the live compose ingestion path so Doctor/knowledge/API corpus is present and searchable.
Completion criteria:
- [x] Compose rebuild reports non-zero knowledge/API/Doctor corpus counts.
- [x] Doctor and release-oriented queries return grounded results where live corpus exists.
- [x] Evidence is recorded in the execution log.
### AI-LIVE-SEARCH-003 - Verify live browser search parity
Status: DONE
Dependency: AI-LIVE-SEARCH-002
Owners: QA, Documentation, Project Manager
Task description:
- Re-run live Playwright verification from `https://stella-ops.local` across Doctor, releases, findings, and policy routes.
- Archive the sprint only after the compose-backed live service behaves consistently with the documented search experience.
Completion criteria:
- [x] Live Playwright evidence shows grounded or explicit supported fallback on the primary routes.
- [x] Documentation reflects the repaired live-search setup.
- [x] Sprint is archived only after all tasks are DONE.
## Execution Log
| Date (UTC) | Update | Owner |
| --- | --- | --- |
| 2026-03-08 | Sprint created after live Playwright reproduction showed that the compose-backed AdvisoryAI service behind `stella-ops.local` lacked knowledge/API/Doctor corpus parity and returned `404` for `POST /v1/search/suggestions/evaluate`. | Developer |
| 2026-03-08 | Root cause confirmed: the compose image published by `stellaops/advisory-ai-web:dev` did not carry the repo-shaped docs/OpenAPI/KnowledgeSearch corpus required by `KnowledgeSearchRepositoryRootResolver`, and the Docker build context excluded the relevant `docs/**` subset. That left the live service with `documentCount=0`, `apiOperationCount=0`, `doctorProjectionCount=0`, and no mapped `POST /v1/search/suggestions/evaluate` surface on the older image. | Developer |
| 2026-03-08 | Repaired runtime packaging by publishing the required corpus assets from `StellaOps.AdvisoryAI.csproj`, allowing the Docker context to include the documented subset via `.dockerignore`, and adding resolver coverage for the published app layout. Focused verification passed: `KnowledgeSearchRepositoryRootResolverTests` `3/3`, `dotnet publish` produced `/app/docs/**`, `/app/devops/compose/openapi_current.json`, and `/app/src/AdvisoryAI/StellaOps.AdvisoryAI/KnowledgeSearch/*.json`. | Developer |
| 2026-03-08 | Rebuilt and force-recreated `stellaops-advisory-ai-web`, then verified live compose parity directly: `POST /v1/search/suggestions/evaluate` returned grounded viability for `database connectivity`; `POST /v1/advisory-ai/index/rebuild` returned `documentCount=470`, `chunkCount=9050`, `apiSpecCount=1`, `apiOperationCount=2190`, `doctorProjectionCount=8`; `POST /v1/search/index/rebuild` returned `domainCount=8`, `chunkCount=8`. | Developer |
| 2026-03-08 | Live Playwright/browser verification on `https://stella-ops.local` passed after the runtime repair. Typed queries returned grounded answers on `/releases/versions` (`database connectivity`, `pending approvals`), `/security/triage` (`critical findings`), `/ops/policy` (`signed SBOM`), and `/ops/operations/doctor` (`database connectivity`). Release-page suggestion chips also executed successfully: `What blocked this promotion?` returned a grounded answer state from the live shell. | Developer |
## Decisions & Risks
- Decision: treat the remaining live-search failure as an AdvisoryAI/runtime parity problem, not as a frontend-only issue.
- Risk: a partial runtime-only hotfix can restore exact searches while leaving generic operator suggestions misleading.
- Mitigation: verify both endpoint parity and corpus parity before claiming live-search recovery.
- Decision: ship the compose parity fix by packaging a bounded repo-shaped search corpus inside the published AdvisoryAI service image rather than relying on host-mounted repo content.
- Decision: keep the live browser proof against `https://stella-ops.local` even after direct backend checks passed, because the original report was a real shell failure.
## Next Checkpoints
- 2026-03-08: confirm compose endpoint parity root cause.
- 2026-03-08: confirm compose corpus parity repair or record blocker.