#1066 fixed ~40 documented calls to methods that do not exist, six pages that began mid-code-block, and 132 identifiers whose _ had been rewritten to *. All three classes were introduced silently by earlier commits and survived months because nothing checks documentation mechanically.
An offline test would catch every one of them, needs no database, and runs in milliseconds:
- fence parity — every
docs/**/*.mdhas an even number of ``` fences. Catches the six truncated pages. - no mangled identifiers — no
[A-Za-z0-9]\*[A-Za-z0-9], and no_between operands. Catches thecreated*at/7 _ 24 _ 60class, while leavingSELECT *,COUNT(*)and**bold**alone. - every
db./trx./tx.receiver names a real method — compare againstObject.keys(createQueryBuilder({ schema })). This is the check that would have caughtdb.update, and it is the cheapest of the four. - every
DB_*env var mentioned in docs is onesrc/actually reads — currentlyDB_USER,DB_TIMEOUT,DB_RETRY_ATTEMPTS,DB_HEALTH_CHECK_INTERVALandDB_WAIT_READY_ATTEMPTSappear in docs and are read nowhere (see #1070).
It fails today on the env-var item even after #1066, so it has immediate value beyond regression cover.
Found while auditing after #1001.