01
Customer and source facts
Purpose, audience, names, rights, and required use come from the customer or source. Inference is labelled as inference rather than presented as fact.
Evidence and limits
Evidence should make limits clearer, not manufacture certainty.
Customers who cannot read the target language should not be asked to infer quality from a score, a confident explanation, or model agreement.
Current service state
No evidence type is silently promoted
Reasoning and model agreement are supporting evidence, not independent proof that a translation is correct. Model reasoning, back-translation, and model agreement are supporting evidence. They are not described as independent proof of correctness.
02
Process and automated evidence
Checks can show what ran, what changed, and what remains. Their evaluator type and limitation travel with the result.
03
Independent or human review
This wording is shown only when an active capability and service policy prove who reviewed what. Missing capability suppresses the claim.
04
Artifact-specific release verdict
Ready, caution, facts required, checking, stale, and do not use are tied to the exact artifact set and current evidence—not just workflow completion.
A strong-looking explanation without an independent evaluator remains self-explanation. leapCAT records that limitation instead of hiding it.