Measure whether your QA graders agree with each other, item by item, before agents appeal their scores.
Free100% in your browserNo sign-up
Runs 100% in your browser. Your files never leave this device.
How it works
Groups graded scorecards by interaction, then compares every grader pair on every scorecard item.
Reports exact-agreement percent, mean absolute difference and weighted Cohen kappa (quadratic weights) per item and overall.
Builds a calibration pack: the 10 highest-variance interactions, with the per-grader scores side by side.
Builds a blind re-grade queue from interactions graded only once, so a second grader can score them without seeing the first grade.
Sample data uses synthetic names only (Agent A, Grader 1). Audio transcription is not in v1; see the good first issues for a future local-only adapter.