Evaluator calibration is the operational answer to the problem that two observers, given the same performance, will score it differently. Without calibration, the scoring system measures the observer as much as the unit. With calibration, the scoring system measures performance.
Calibration in OCTAAR runs continuously, not as a one-time exercise. Per-evaluator variance is reported against the calibrated baseline; out-of-tolerance scoring is routed back for review before it shapes a finding. Drift is detected, surfaced, and trained out — not absorbed.