Evidence & freshness

Capabilities and prices remain useful only when you can tell where they came from, how they were verified, and when they were observed.

Capability status

supported

The deployment supports the capability under the recorded conditions.

conditional

Support exists only when the listed conditions are met.

unsupported

Current evidence indicates the capability is not supported.

unknown

There is not enough current evidence to make a claim.

not_applicable

The capability does not apply to this deployment mode.

Verification status

official_declared

Declared in first-party provider documentation.

cross_checked

Confirmed by more than one independent source.

probed_pass / probed_fail

Observed in a controlled runtime probe.

not_probed_budget

A probe was intentionally skipped because it would incur cost.

unknown

No qualifying verification is available.

Freshness fields

observed_at records when a deployment or rate was last observed. source_url and evidence_urls make the claim auditable. The response header x-allm-catalog-release identifies the immutable snapshot that served it.

Production policy

For critical workflows, require both an acceptable capability status and verification level. Decide how old evidence may be, reject unknown, and re-evaluate when the catalog release changes.

{
  "requires": { "tool_calling": "supported" },
  "verification": ["official_declared", "cross_checked", "probed_pass"],
  "max_evidence_age_days": 30,
  "on_unknown": "reject"
}