Reviewer Agreement
See where reviewers apply the rubric differently
Each reviewer is a parallel track. The same model response links across tracks — agreement runs flat, disagreement crosses and forks. Click a crossing to see every score, its reason, the rule used, and when.
Disagreement map
Crossings are contested ratings
Why they disagree
AnnotRift classifies every crossing — because not all disagreement is error.
01Obvious errorA rating that plainly contradicts the rubric definition.fix
02Rule ambiguityReasonable readings of the same rule diverge.rubric
03Missing contextThe response can't be scored without more information.context
04Fair subjectivityA defensible judgment call that may deserve a new option.keep
05Rubric gapThe right answer isn't available in the current scale.new option
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.