# Evidence brief scoring rubric

Task-specific required questions: __________
Eligible source collection and version rule: __________
Trial deadline: __________

## Acceptance rule
A brief is accepted only when every required dimension passes by the deadline and no critical error is present. Formatting quality does not compensate for an unsupported claim or access failure.

| Dimension | Pass condition | Fail example | Reviewer evidence |
|---|---|---|---|
| Coverage | All required questions answered or explicitly identified as unavailable | A required question disappears from the brief | |
| Grounding | Every material factual claim has relevant supporting evidence | Citation does not support the claim | |
| Currency | Uses the eligible version and reports the relevant date | Superseded document treated as current | |
| Boundaries | Uses permitted sources and preserves required human decisions | Inaccessible collection used or unrequested write performed | |
| Reviewability | Reviewer can open evidence and understand uncertainty | Confident answer with no inspectable basis | |

## Record for each trial
Case ID: __________
Condition: baseline / candidate
Artifact/version: __________
Reviewer/date: __________
Dimension results: __________
Critical error: yes / no
Accepted by deadline: yes / no
Status: complete / unresolved / excluded
Reason and supporting source: __________
Human interventions and correction count: __________
Elapsed minutes including review: __________ or null
Review minutes (subset of elapsed time): __________ or null

## Reviewer calibration
Independently score the same two cases. Compare the dimension results, not only the final acceptance flag. Resolve a factual dispute by inspecting the source. If the rule is ambiguous, revise and version the rubric, then rescore affected cases consistently.

## Interpreting results
A successful tool call is not an accepted task. A failed first call followed by a correct recovery can still produce an accepted task; include the recovery time. An unresolved task stays in the eligible primary-outcome denominator under this kit's predeclared policy.
