HealthPY Pro — Standalone

## The scenario The Activity team has asked you to **productionise** the monthly reconciliation work into something an oncall analyst can rely on: scheduled, observable, and able to fail gracefully without dropping a month's commissioning view. ## Deliverables A Python project structured as a real service: 1. **Input validation** with pydantic (or attrs + cattrs) for the four CSV schemas. Bad files reject cleanly with a structured error report. 2. **Idempotent monthly run** — replay any month without producing duplicate spell records. 3. **Structured logs** (JSON Lines via stdlib `logging` is fine) with run-id correlation across stages. 4. **A small HTTP surface** (FastAPI or `http.server` — your call) exposing `/runs/latest`, `/runs/{id}`, `/health`. No frontend. 5. **A scheduled-run script** (`cron`-style or APScheduler) that picks up the next month when its files land. 6. **Tests** with pytest: at least the lookup, the audit-flag rules, and the idempotency guarantee, with fixtures. 7. **A README and an OPS runbook** — "the file didn't arrive, what do I do?" / "a row failed validation, where do I look?" ## Acceptance criteria (summary) pydantic-style schema validation · idempotent re-runs · structured logs with run-id · pytest suite passing locally · README + OPS runbook · containerised (Dockerfile + compose) · ≥15 conventional commits. Full brief, dataset orientation, and architecture rubric appear inside the lesson once enrolled.