4.0: every value list is a closed enum or an open vocabulary, with one extension-term schema, checked by the build #9002

Revised 2026-11-14: every enum takes the open form, except discriminators; x-closed is dropped.

Problem

A value list whose receivers must accept unknown values isn't an enum, yet the schemas declare every such list with enum, which a strict validator enforces. So a conforming strict receiver rejects what the versioning page tells it to accept:

Clients must tolerate unknown values and fall back to sensible defaults. (versioning.mdx L228)

With #8915, a 3.3 adagents.json using the new match_keys value full_name fails 3.2.10 validation as a whole file. advertiser-industry.json says "Sellers MUST accept unknown values", and its own enum rejects them.

In static/schemas/source at main bf4aa6dbb3:

This gist has every site, the citations and the scripts.

Proposal for 4.0

1. One form for every list. Every enum takes the open form:

   {"oneOf": [{"enum": ["automotive", "finance"]},
              {"$ref": "/schemas/core/extension-term.json"}]}

   A minor adds core values, and anyone may send an extension term. Only discriminators differ: a one-value enum becomes const, and a branch-selecting key stays plain.
2. One extension-term schema. A term is {domain}:{key}, on a domain its definer controls, as in BrandRef, so no registry is needed. Core values never contain :, so oneOf is exact; chroma_subsampling is respelled.
3. No exemption. The tolerate-unknown text goes: a sender uses a core value only with peers served a release that defines it, as with fields.
4. Unknown terms. Each open position's x-adcp-unknown-term is ignore or the core value an unknown term counts as.
5. One lint. Every enum is a discriminator or an open-form core, and every extension branch references the shared schema. Its 4.0.0 baseline is empty.

Why 4.0

Moving custom:, x_ and the URI branches to the shared grammar changes wire values, and dropping the exemption changes what sellers may send. It fits T2 beside #8900 and #8901. For #8885, error.code takes the open form.

I'm happy to take this on.

Analyzed and generated with the help of Claude, fully reviewed by me.