Ship sprint stories end-to-end

You are implementing E19/E20 sprint stories for ACM-AI — an intelligent Asbestos Containing Material compliance management system. Read `.ralph/@fix_plan.md` for the current task list. Read `CLAUDE.md` for project conventions and architecture. Read `docs/sprint-artifacts/e19-e20-implementation-prompts.md` for per-story detailed implementation instructions. ## Workflow Per Iteration 1. Read `.ralph/@fix_plan.md` — find the next unchecked task `- [ ]` 2. Read the corresponding story file from `docs/sprint-artifacts/` (path is in the task line) 3. Read the matching story section in `docs/sprint-artifacts/e19-e20-implementation-prompts.md` for detailed implementation steps 4. Implement the story following patterns in `CLAUDE.md` 5. Delegate to specialists based on file scope: - `migrations/`, `api/`, `open_notebook/`, `commands/` → `backend-specialist` - `frontend/` → `frontend-specialist` - Mixed stories (backend + frontend): delegate to `backend-specialist` FIRST, then `frontend-specialist` (API before UI) - Only use specialists when the story touches multiple files/layers; implement simple single-file changes directly 6. Run full verification after all implementation: ``` uv run ruff check . uv run pytest tests/ -x cd frontend && npm run lint && npm run build ``` 7. If all verification passes: - Check off the task in `.ralph/@fix_plan.md` (change `- [ ]` to `- [x]`) - Mark story `done` in `docs/sprint-artifacts/sprint-status.yaml` - Append a Dev Agent Record to the bottom of the story file: ``` --- ## Dev Agent Record **Implemented:** [date] **Files changed:** [list key files] **Tests added:** [list test files] **Verification:** ruff ✓ | pytest ✓ | lint ✓ | build ✓ ``` - Advance the NEXT story in the sequence to `ready-for-dev` in `sprint-status.yaml` (see sequence below) - Log completion to `docs/sprint-artifacts/party-mode-20260224/progress.md` (create if missing) - Commit: `feat(STORY-ID): brief description of what was built` 8. If verification fails: fix and retry (max 3 attempts), then output `<promise>BLOCKED</promise>: [reason + error details]` 9. After ALL tasks are `[x]` in `@fix_plan.md` and verification passes: output `<promise>COMPLETE</promise>` ## Story Sequence (STRICT ORDER — one at a time) ``` E19-S1 → E19-S2 → E19-S3 → E19-S4 → E19-S5 → E19-S6 → E19-S7 → E19-S8 (P1) → E20-S1 → E20-S2 → E20-S3 → E20-S4 ``` When advancing the next story, use this map: - After E19-S1 done → set `e19-s2-jobs-dashboard: ready-for-dev` - After E19-S2 done → set `e19-s3-feature-gating: ready-for-dev` - After E19-S3 done → set `e19-s4-raw-extraction-table: ready-for-dev` - After E19-S4 done → set `e19-s5-building-review-wizard: ready-for-dev` - After E19-S5 done → set `e19-s6-acm-schema-mapping-wizard: ready-for-dev` - After E19-S6 done → set `e19-s7-job-detail-page: ready-for-dev` - After E19-S7 done → set `e19-s8-conversational-crud-chat: ready-for-dev` - After E19-S8 done → set `e20-s1-page-boundary-fix: ready-for-dev` - After E20-S1 done → set `e20-s2-regex-yield-check: ready-for-dev` - After E20-S2 done → set `e20-s3-not-sampled-capture: ready-for-dev` - After E20-S3 done → set `e20-s4-e2e-accuracy-validation: ready-for-dev` ## Special Rules ⚠️ **E19-S1 DESTRUCTIVE**: Migration 032 deletes ALL `acm_record` rows and adds `review_status` to `source`. This is expected and approved by the user — proceed without additional confirmation. ⚠️ **E20-S4 GATE**: Do NOT run real PDF extraction for E20-S4 until E20-S1, E20-S2, AND E20-S3 unit tests all pass. Before the extraction run, verify: ``` uv run pytest tests/test_building_inventory*.py tests/test_orchestrator*.py tests/test_acm_extraction*.py -x ``` Only run ONE real extraction on `docs/samplePDF/` to validate. Never re-run unless a specific bug is confirmed unfixed. 💰 **COST AWARENESS**: Never trigger real LLM extraction calls during unit test development. Write all code and tests first, l