Data Bias Audit & Mitigation Plan

Conduct a data bias audit for {{dataset}}:

1. Bias sources:
   - Historical bias (training on past inequitable data)
   - Sampling bias (underrepresented groups)
   - Measurement bias (proxy variables for protected attributes)
   - Aggregation bias (one-size-fits-all model)

2. Fairness metrics:
   - Demographic parity (equal rates across groups)
   - Equalized odds (error rates balanced across groups)
   - Calibration (predicted probabilities match observed outcomes by group)
   - Choose 2-3 metrics relevant to {{use_case}}

3. Baseline assessment:
   - Current performance by protected attribute (race, gender, age, etc.)
   - Quantify disparity (% point difference, ratio)

4. Mitigation strategies:
   - Collect more balanced data for underrepresented groups
   - Re-weight training data
   - Apply fairness constraints during model training
   - Explainability / transparency improvements

5. Governance:
   - Who owns fairness monitoring?
   - Audit cadence (quarterly, annually?)
   - Escalation if fairness thresholds breached

Deliverables: bias assessment scorecard + prioritized action plan.