peopleanalyst

Platform · Building · new here

Single-model fairness monitor (automated-decision disparity audit)

ONE classifier trained with a fairness-aware loss (demographic-parity / equalized-odds penalty) plus an external held-out audit cohort that measures protected-class disparity post-hoc. Non-infringing design-around of US11922435B2's dual-model bias architecture: no second/monitoring model, no pseudo-unbiased corpus, no dual-classifier comparison. Complements the HO-191 pay-equity analyzer (compensation disparity) on the protected-class fairness thread.

Cluster
Platform
Type
algo
Status
Building
Used by
people-analytics-toolbox

Maturity evidence

Functioning
train + audit implemented (core/fairness-monitor.ts); 8/8 tests pass (tests/fairness-monitor.test.ts): biased cohort flags disparity, fairness-loss training reduces measured disparity, degenerate single-group case, deterministic, structural boundary.
Valuable
Complements the HO-191 pay-equity analyzer on the protected-class fairness thread (decision-outcome disparity vs compensation disparity). No paying consumer yet; in-force design-around pending legal review before commercial use.
Understood
README section + mermaid (train-with-fairness-loss -> audit-cohort -> disparity-report) + machine-readable boundaryAttestation in every report.
Integrated
REST + MCP from one declaration (registered in wage-benchmark MCP register + registry.ts contracts). No downstream consumer wired yet.

How to use it

REST
/api/spokes/wage-benchmark/fairness-monitor/train, /api/spokes/wage-benchmark/fairness-monitor/audit
MCP
wage-benchmark.fairness-monitor.train, wage-benchmark.fairness-monitor.audit
Data
src/spokes/wage-benchmark/core/fairness-monitor.ts

Other Platform capabilities

← Every platform capability

Capability detail sourced from the People Analytics Toolbox capability feed (source of truth), snapshot retrieved 2026-07-23.