Engine calibration receipts

Once per UTC day, each engine receives the probe query: "what is wikipedia". This is an unambiguous question every working engine must answer correctly — the reference to wikipedia.org must appear in the cited sources. If it doesn't, the parser is producing garbage results.

A failed probe means the engine is immediately marked unhealthy and excluded from all user citation checks until it passes the next day's probe on its own — no operator action required. We never serve results from a drifted parser.

Engines that can drift silently are canaried every day. Engines that fail loudly surface hard HTTP errors instead — those failures are visible as engine-level errors in your results.

Citation-tracking tools have been documented reporting materially false citation rates — engines returning stale cache, parser bugs mistaking related content for citations, or headless browser quirks producing random false positives. None of those tools show their work.

Seeklyn's honesty moat is built on one principle: if you can't explain the methodology, the number means nothing. Publishing the calibration receipts lets you verify that the engines we queried for your check were actually working that day.

Using another citation tracker? Ask them for their calibration receipt. Ask which engines they run, how they detect a citation versus a mention, what happens when an engine's parser drifts, and how they know when results are stale. If they can't answer, you're trusting an unverifiable number.