CrucibleIR.Reliability.Fairness (CrucibleIR v0.2.1)

Configuration for fairness and bias detection.

Controls fairness metrics, group definitions, and violation handling.

Fields

:enabled - Whether fairness checking is enabled (default: false)
:metrics - List of fairness metrics to compute
:group_by - Attribute to group by for fairness analysis
:threshold - Fairness threshold (e.g., 0.8 for 80% rule)
:fail_on_violation - Whether to fail when violations detected
:options - Additional fairness options

Available Metrics

:demographic_parity - Equal positive prediction rates
:equalized_odds - Equal TPR and FPR across groups
:equal_opportunity - Equal TPR for qualified candidates
:predictive_parity - Equal positive predictive values

Summary

Types

metric()

t()

Types

metric()

@type metric() ::
  :demographic_parity
  | :equalized_odds
  | :equal_opportunity
  | :predictive_parity
  | atom()

t()

@type t() :: %CrucibleIR.Reliability.Fairness{
  enabled: boolean(),
  fail_on_violation: boolean() | nil,
  group_by: atom() | nil,
  metrics: [metric()] | nil,
  options: map() | nil,
  threshold: float() | nil
}