CalibrationBias

Cost asymmetry between false negatives and false positives.

bias_direction is always auto-derived — never set by the caller: false_negative_cost > false_positive_cost → recall (missing a real event is more costly; prefer sensitivity) false_positive_cost > false_negative_cost → precision (spurious alerts are more costly; prefer specificity) equal costs → balanced

false_negative_costFalse Negative Cost (string)required

Possible values: [high, medium, low]

false_positive_costFalse Positive Cost (string)required

Possible values: [high, medium, low]

bias_directionBias Direction (string)

Possible values: [recall, precision, balanced]

Default value: balanced

CalibrationBias

{
  "false_negative_cost": "high",
  "false_positive_cost": "high",
  "bias_direction": "balanced"
}