Why passing a code check and recording the truth are not the same goal

A developer running a semantic checker on a codebase encountered 11 unrecognised identifiers, which could apparently be resolved by adding all 11 entries to a single vocabulary file. On closer inspection, the identifiers belonged to two distinct categories — six module-level instruments and five layer-level symbols — each requiring a separate registry with different meanings and downstream readers. Bulk-declaring all entries in one place would have cleared the check but silently introduced false data, misleading other tooling about what operations a module actually performs. Even after the correct two-file fix was made, the checker continued reporting errors because it reads a generated artifact rather than the source declarations directly, requiring a commit-then-regenerate cycle before results reflect reality. The author concludes that a passing check and an accurate codebase are different goals, and when they conflict, the check — as a downstream artifact — should be treated as subordinate to the truth.