Verification log — Webhook-to-CRM Sync Microservice (Zapier Replacement)
Project: ~/freelance-engine/projects/webhook-crm-sync/
Verified by Clancy, 2026-11-12 / 2026-11-13 UTC. Gemini wrote the code (Pro/Medium); Clancy audited, fixed, executed, and verified it.

TEST SUITE — 6/6 PASSED
  pytest tests/test_app.py -v → 6 passed (1 warning: Starlette TestClient
  deprecation notice about httpx — cosmetic, not a failure).
  Covered: valid submission → 201 synced + CRM called once + log row;
  invalid email → 422; missing required field → 422; duplicate email →
  200 duplicate + CRM called only once; downstream failure → 502 + row
  logged failed with crm_id null; GET /health → 200 {"status": "ok"}.

FIXES APPLIED BY CLANCY (marked in the files)
  1. src/app.py — added conn.commit() after the failed-row INSERT, before
     raising the 502. Without it, the HTTPException unwound the sqlite
     `with` block and the failed row rolled back, so downstream failures
     were never logged (the brief requires them to be). This was a real
     bug: the downstream-failure test failed (0 rows in the log) until
     the fix.
  2. tests/test_app.py — replaced AsyncMock return-value attribute setup
     (mock_post.return_value.json.return_value = ...) with real
     httpx.Response objects including an attached request. The mock
     setup made .json() and .raise_for_status() async, so resp.json()
     returned an un-awaited coroutine (AttributeError: 'coroutine'
     object has no attribute 'get'). Real Response objects fixed it.
  3. run_demo.sh — exports a clean NO_PROXY ("localhost,127.0.0.1").
     This VM's NO_PROXY contains bracketed IPv6 literals that httpx
     cannot parse (InvalidURL: Invalid port: ':1]'), which broke every
     test that constructs an httpx client. Same class of VM-egress
     quirk as the Playwright relay patch.
  4. README.md — fixed the test command (it pointed at
     tests/verify_deliverable.py; the suite is tests/test_app.py) and
     closed an unclosed ```bash code fence.
  5. Environment — installed email-validator in the project venv
     (pydantic[email] requires it; it was not installed).

LIVE DEMO (run_demo.sh) — PASS ON RE-RUN
  Run 1: both live submissions returned 502 "Downstream CRM error: "
  (empty detail). The failed rows WERE logged to SQLite (proving fix 1
  live). Cause was transient: cold egress-proxy connections timing out
  against httpx's 5s default on the first-ever outbound call from the
  fresh uvicorn process (run took ~13s vs ~3s on re-run).
  Run 2 (re-run, same script, no code changes): full pass.
    - GET /health → {"status": "ok"}
    - POST /webhooks/contact (live@example.com) → 201
      {"status": "synced", "crm_id": "11", "contact_id": 1}
    - POST same email again → 200 {"status": "duplicate", "contact_id": 1}
    - GET /sync-log → exactly 1 row (live@example.com, crm_id 11, synced)
  Independent check (read demo_sync.db directly with sqlite3):
    rows: [(1, 'live@example.com', '11', 'synced')] — PASS.

CONCLUSION
  The microservice works as briefed: validation, dedupe, CRM forward,
  failure logging, and the sync log all verified live against the real
  JSONPlaceholder endpoint. The one transient failure was environmental
  (cold proxy), not a code defect, and the re-run passed clean.