Elicited Adaptation: Auditable Localized Fairness via Pairwise Queries

A single global fairness criterion cannot serve communities whose conceptions of fair treatment differ. We introduce elicited localized fairness: an end-to-end pipeline that elicits per-community Mahalanobis metric–tolerance specifications ${\text{cF}}_k=(d_k,{\text{eps}}_k)$ from stakeholders via two phases of pairwise queries (similarity, then tolerance acceptability), trains a localized classifier (LocAdapt), and audits the deployed model on a held-out tolerance fold. Theory. A constrained-MLE rate $\tilde{O}(\sqrt{d^2/n})$ for metric elicitation and a matching $\Omega (d^2/t^2)$ active-query lower bound, exact in $(d,n,t)$ exponents; a held-out audit certificate that converts Phase 2 queries into a PAC-style violation guarantee. Empirics. On heterogeneous-community settings (Stress $K=5$, Adult-K3-human; 10 seeds), LocAdapt sits on the (accuracy, strict $d\le 1$ violation) Pareto frontier: averaged and matched-accuracy Global IF lose on strict at comparable accuracy ($40$–$80\%$ higher), and pooled-MLE Global IF matches strict only at $-2$pp accuracy. A five-axis diagnostic (acc, strict, score-diff spread, prediction entropy, AUC) confirms the gain is not prediction-smoothing. Plugging Phase 2 elicited tolerances into LocAdapt training reduces the COMPAS deployed-spec violation $66\%$ ($p\ll 10^{-4}$). Five pre-registered human studies on COMPAS (360 annotators total) validate the pipeline end-to-end; a 180-annotator scorer-stability ablation across three scorer architectures preserves the cross-community $\widehat{\text{eps}}$ ordering and pools to $p=0.0014$ ($n=90$ vs. $90$).