D119 - A supplied name is not a published one

decided · 2026-08-19 · found while designing the probe pipeline

--words names were merged into the published-standard list and the whole batch recorded as published-standard, citing a file they had never been in.

It failed safe - an audit checks membership in that file, does not find them, and reports them unaccounted - but it reported the wrong problem, and the sharper issue was this: the Supplied variant was unreachable. D073 defined the category whose entire job is labelling outside material, and no code path could produce one. The mechanism for being honest about imported names did not work.

The two lists are now searched separately, because they are different kinds of claim, and --words-from says which:

• observed - learned by running something we wrote. Ours, but not reproducible by re-running an index, so an audit lists it separately rather than counting it verified.
• supplied - came from outside. Never verifies. The default, because assuming the more generous label is exactly the mistake an audit exists to prevent.

D213 replaced the first of those with probe, which says what the list actually is: names our own conformance probe reported, on hardware. Same distinction, named after its source rather than after the fact that something ran.