{
  "$schema": "model-registry/1.0.0",
  "$comment": "Which languages a model is good at, and whether anybody checked. On the owner's instruction: 'should clarify each new model added to bootstrap which of the preferred languages ... especially if human validated'. NO LANGUAGE CLAIM IS MADE HERE BY AN AGENT, and the absence of one is reported rather than silent — the harness that reads this file names it as a finding at every run. Two reasons it ships empty. First, an agent is not a valid source: a model's own claim about which languages it handles well is a generated assertion about a generated system, which is the exact thing `validation` exists to distrust, so an agent populating this would be manufacturing the evidence the schema asks for. Only `human-validated` is ever acted on, and only a person can grant it. Second, a language claim about a model is a claim about a third party's system, and writing one nobody checked would be worse than the fallback it replaces. Add an entry when a person has actually looked: id, title, preferredLanguages (an EMPTY list is a determined empty — somebody looked and found none notable; `validation: \"unverified\"` is nobody having looked), validation, and validatedBy + validatedOn when human-validated. Schema: schemas/model-registry.schema.json. How the languages are USED is the reading harness's business: as one input, never the answer. IDENTITY-ONLY ENTRIES (#1168 B10c, owner 2026-09-30: 'resolve properly'): an `unverified` entry with NO `preferredLanguages` records only that a model id exists, so a record's `model` field resolves to a node here; it asserts nothing about languages and is never read by the language determination. Such an entry may be added for any id the corpus records (check-model-languages lists the unregistered ones). A LANGUAGE claim — adding `preferredLanguages`, or moving to `self-reported` or `human-validated` — stays a person's act.",
  "models": [
    {
      "id": "claude-opus-5",
      "title": "claude-opus-5",
      "validation": "unverified",
      "note": "Identity only: this id is recorded by the corpus (model fields in QA sidecars, attributions and adjudications). Its languages have not been examined, so it carries no preferredLanguages and makes no language claim."
    },
    {
      "id": "claude-opus-5-5",
      "title": "claude-opus-5-5",
      "validation": "unverified",
      "note": "Identity only: this id is recorded by the corpus (model fields in QA sidecars, attributions and adjudications). Its languages have not been examined, so it carries no preferredLanguages and makes no language claim."
    },
    {
      "id": "claude-sonnet-5",
      "title": "claude-sonnet-5",
      "validation": "unverified",
      "note": "Identity only: this id is recorded by the corpus (model fields in QA sidecars, attributions and adjudications). Its languages have not been examined, so it carries no preferredLanguages and makes no language claim."
    },
    {
      "id": "gemini-1.5-pro",
      "title": "gemini-1.5-pro",
      "validation": "unverified",
      "note": "Identity only: this id is recorded by the corpus (model fields in QA sidecars, attributions and adjudications). Its languages have not been examined, so it carries no preferredLanguages and makes no language claim."
    },
    {
      "id": "gemini-2.5-pro",
      "title": "gemini-2.5-pro",
      "validation": "unverified",
      "note": "Identity only: this id is recorded by the corpus (model fields in QA sidecars, attributions and adjudications). Its languages have not been examined, so it carries no preferredLanguages and makes no language claim."
    }
  ]
}