You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Every prior ticket verified its own slice. Nothing yet verifies the integrated whole: clean clone -> sync -> all gates -> running stack -> green CI.
Proposed solution
Run the full verification checklist from the plan: clean-clone smoke, uv sync && just check, cd frontend && npm install && npm run lint && npm run check && npm run test && npm run build, pre-commit run --all-files, docker compose up and hit all three services, open a sample PR and confirm every CI gate passes, trigger eval-nightly via workflow_dispatch with sample secrets and confirm execution. File any defects discovered as bug issues against the relevant prior ticket.
Acceptance criteria
All checklist items in the plan's "Verification checklist" tick green.
No file mentions "Teller", "financial", or other Teller-specific identifiers (grep verified).
Sample PR closed without merging (verification only); summary of green checks attached.
Priority rationale
Critical: this is the gate that decides whether to publish or hold.
uv run pre-commit run --all-files: ✅ (one Windows-side CRLF/LF cosmetic difference; no content drift, CI is unaffected)
cd frontend && npm ci: ✅ (153 deps, 0 vulnerabilities)
cd frontend && npm run lint: ✅ (0 warnings, max-warnings=0)
cd frontend && npm run check: ✅ (tsc --noEmit clean)
cd frontend && npm run format:check: ✅
cd frontend && npm run test: ✅ (2 passed)
cd frontend && npm run build: ✅ (Vite production bundle ~190 KB / 60 KB gzipped)
Forbidden-token sweep
grep -ri "teller|financial|tomoro|takehome|duckdb" — zero matches across the source tree (cleanup landed in #62).
What's intentionally not verified here
docker compose up end-to-end stack boot — local Docker Desktop wasn't available during this run; the Dockerfile + compose file shape is reviewed and CI's Container image scan (trivy) builds the same image successfully every PR.
eval-nightlyworkflow_dispatch smoke — would need real LLM secrets configured; the workflow file shape is reviewed and the local pytest eval/ exercises the same code path with the placeholder _answer_fn.
Branch-protection apply workflow — needs the BRANCH_PROTECTION_TOKEN PAT secret; documented in docs/DEVELOPMENT.md#branch-protection-sync-setup.
These three items are documented next-steps for #28's publish phase.
Activity
chore: scrub remaining Teller / financial / DuckDB references (#27) (#62
constk commentedon Apr 27, 2026
Verification summary
Latest develop (
707f49d) — all gates green.CI (
.github/workflows/ci.yml) — 10 jobsSecurity (
.github/workflows/security.yml) — 4 jobs.github/security/pip-audit-ignore.txtwith sunset note)Local verification
uv sync --extra dev: ✅uv run ruff check . && ruff format --check .: ✅uv run mypy --strict src/ tests/: ✅ (31 source files)uv run lint-imports: ✅ (2 contracts kept)uv run pytest tests/ --cov=src: ✅ (40 passed, 98.64 % coverage)uv run pytest eval/: ✅ (1 caseecho-hellopassed)uv run pre-commit run --all-files: ✅ (one Windows-side CRLF/LF cosmetic difference; no content drift, CI is unaffected)cd frontend && npm ci: ✅ (153 deps, 0 vulnerabilities)cd frontend && npm run lint: ✅ (0 warnings, max-warnings=0)cd frontend && npm run check: ✅ (tsc --noEmit clean)cd frontend && npm run format:check: ✅cd frontend && npm run test: ✅ (2 passed)cd frontend && npm run build: ✅ (Vite production bundle ~190 KB / 60 KB gzipped)Forbidden-token sweep
grep -ri "teller|financial|tomoro|takehome|duckdb"— zero matches across the source tree (cleanup landed in #62).What's intentionally not verified here
docker compose upend-to-end stack boot — local Docker Desktop wasn't available during this run; the Dockerfile + compose file shape is reviewed and CI'sContainer image scan (trivy)builds the same image successfully every PR.eval-nightlyworkflow_dispatchsmoke — would need real LLM secrets configured; the workflow file shape is reviewed and the localpytest eval/exercises the same code path with the placeholder_answer_fn.BRANCH_PROTECTION_TOKENPAT secret; documented indocs/DEVELOPMENT.md#branch-protection-sync-setup.These three items are documented next-steps for #28's publish phase.