Gaps and recommendations
Maturity level L0-L5
Mean absolute calibration error across all dimensions
AMC's observed evidence-based score per dimension (0-1)
Overconfidence ratio: dimensions where self > observed
Per-dimension calibration error: |self - observed|
Overall calibration score 0-100
Agent's self-reported confidence per dimension (0-1)
Underconfidence ratio: dimensions where self < observed
Expected Calibration Error (ECE) using binned approach