Primary objects
review run
One submitted AI output plus supplied sources, intended use, metadata, processing state, and review artifacts.
atomic claim
A factual, normative, causal, recommendation, or citation claim extracted from the submitted output.
source span
A quoted evidence span from a supplied source, with offsets when available.
verification result
Stored claim-level source-support verification metadata and support status for one atomic claim against candidate source spans.
claim report
The reviewer-facing artifact summarizing unsupported claims, citation issues, evidence notes, and follow-up questions.
gate verdict
Machine-readable workflow decision: allow, allow_with_limits, needs_review, needs_repair, or blocked.
Support status
Support status is claim-level. It describes whether a specific atomic claim appears supported by the supplied sources available to the verifier.
| Status | Meaning | Caller action |
|---|---|---|
supported | The claim appears supported by supplied source evidence. | Usually continue. |
partially_supported | The source supports part of the claim or narrower wording. | Review or edit before use. |
unsupported | No supplied source supports the claim. | Repair the output or route to review. |
contradicted | A supplied source appears to conflict with the claim. | Stop workflow use until repaired. |
insufficient_context | The supplied sources do not provide enough context. | Add sources or route to review. |
not_checked | Verification was not completed for the claim. | Poll, retry, or inspect failure state. |
Use status and gate verdict
The `safe_use_status` API field summarizes the source-support review outcome for the submitted intended use. Gate verdict is the machine-readable mapping for workflows that need a compact action signal.
| Safe-use status | Meaning | Gate verdict |
|---|---|---|
verified_for_use | No unsupported or blocking issues were found in the supplied-source review. | allow |
verified_with_limits | The claim report contains limits, caveats, or partial support. | allow_with_limits |
needs_review | Human attention is recommended before use. | needs_review |
needs_repair | The output should be changed or rerun before use. | needs_repair |
blocked | The workflow should not use the output in its current state. | blocked |