Proof
Test counts and defect notes, checked against the artefacts. If the live page disagrees with the test suite, the live page wins.
| Artefact | Tests | Date checked |
|---|---|---|
| URL and column-spec scraper | 6 passed | 2026-11-19 |
| Webhook receiver with dedupe and sync log | 5 passed | 2026-11-19 |
| Playwright smoke-test harness | 4 passed | 2026-11-19 |
| OpenAPI skeleton | 5 passed | 2026-11-19 |
| HN front-page scraper | 6 of 6 passed | 2026-11-12 live, rechecked 2026-11-19 |
20 tests passed, 0 failed across the four bench artefacts, run from a clean shell on 2026-11-19
Defect notes
HN scraper: green tests, empty submitter column
The submitter selector relied on the .hnuser class, which is no longer present on live Hacker News. The mocked fixture still carried the old class, so the suite passed while every row wrote an empty submitter. Caught by checking the live page. Fixed with a[href^="user?id="] and tests that fail when the selector stops matching. 30 rows, submitter populated on every row after the fix.
Verified 2026-11-19
Webhook receiver: duplicate deliveries
A repeat delivery of a synced event returns a duplicate result without reprocessing, proven by a counting processor in the tests. A failed event retries on redelivery and lands at synced with attempts 2. Invalid payloads return 422 with no log rows.
Verified 2026-11-19
OpenAPI skeleton: spec and app held together
The conformance test fetches /openapi.json from the app and asserts every YAML path and method is served, the schemas exist, and each error response uses the one Error schema. Validation ran through structural asserts because openapi_spec_validator is not in the shared venv.
Verified 2026-11-19
Every suite runs offline against bundled fixtures or in-memory stores. Nothing here needs a paid API, a secret, or a client account. Raw files: pages.milehighwebplug.xyz