Audit & Evidence Board
Fairness Testing
Evaluating individual fairness by simulating single-attribute demographic flips under fixed financial parameters.
Individual Fairness Twin Simulation
Perturbation testing: Evaluating predicted default probability when flipping IsMinorityOrWomanOwned while holding cash flow and credit attributes identical.
Profile A: Observed MWBE Enterprise
IsMwbe = 1
11.3% PD
Automated Scoring Engine Default Probability
Profile B: Counterfactual Twin
IsMwbe = 0
9.5% PD
Simulated Default Probability (Demographic Flip)
| Underwriting Attribute | Profile A (Observed) | Profile B (Counterfactual) | Attribute Variance (Δ) | Fairness Impact |
|---|---|---|---|---|
| MWBE Protected Group Flag | Protected (True) | Control (False) | 1-Bit Flip | Simulated Perturbation |
| Commercial Bureau Score (Paydex) | 67 | 67 | 0.0 (Identical) | Controlled |
| Debt Service Coverage Ratio (DSCR) | 1.36 x | 1.36 x | 0.0 (Identical) | Controlled |
| Annual Commercial Revenue | $633K | $633K | 0.0 (Identical) | Controlled |
| Engine Predicted Probability of Default (PD) | 11.3% | 9.5% | +1.8% Gap | NOMINAL |
Individual Fairness Audit Finding:
Individual counterfactual parity is satisfied with a negligible delta of +1.8%, indicating that the scoring model does not penalize protected demographic indicators.
Individual counterfactual parity is satisfied with a negligible delta of +1.8%, indicating that the scoring model does not penalize protected demographic indicators.