Dataset Reports

Explain evaluation quality beyond a single score

Not a dashboard — an audit journal. Each conclusion carries real data and the records behind it. Produce a team-detailed version, a management brief, or a data-supplier delivery report, with AI findings, human-confirmed items, and open issues clearly separated.

Evaluation Reliability

Open records →
verify

Overall agreement after rubric v13, weighted by criterion. Confirmed by expert review on contested pockets.

91%agreement
912verified
v13rubric

Reviewer Differences

Open records →
conflict

Concentration of disagreement by reviewer pair and criterion. Accuracy drives most contested ratings.

1,364rifts
19pairs
Accuracyhot criterion

Rubric Ambiguity

Open records →
analysis

Rules flagged by the Ambiguity Lens with the disagreement each one explains.

3ambiguous
64%explained
1fix queued

Edge Case Coverage

Open records →
evidence

Coverage of known failure modes by the current evaluation set and gold references.

24modes
19covered
5gaps

Gold Set Status

Open records →
verify

Size, version, and requirements met by the current trusted reference dataset.

912gold
v12version
0leakage

Recommended Actions

Open records →
muted

Prioritized next steps — separated into AI-suggested, human-confirmed, and still-open.

7suggested
4confirmed
3open
Evaluation Reliability Layer

Find what makes your AI evaluations unreliable.

Point AnnotRift at your evaluation set and get a traceable quality report in minutes — every finding backed by data and rules, every decision confirmed by a human.