Explainers: Adverse Impact at Each Hiring Stage, Compounding Disparities, and Differential Validity #1119

The AI Fair Recruitment audit measures one decision, but hiring is a pipeline of decisions, and Disparate Impact is usually computed for the last one. Add all three in one pull request.

explainers/hiring-pipeline-adverse-impact.md - How Do You Measure Adverse Impact at Each Stage of a Hiring Pipeline?

Selection rate and the four-fifths ratio per stage (screen, test, interview, offer) and for the whole funnel. Build the stages for a real audit from its columns (for example test score and interview score thresholds), report the ratio and an interval at each stage, and show which stage contributes most. Keep legal statements to primary sources.

Key citation to verify: The Uniform Guidelines on Employee Selection Procedures (29 CFR 1607) for the four-fifths rule, and Bogen, Rieke (2018), "Help Wanted: An Examination of Hiring Algorithms, Equity, and Bias" (Upturn) - confirm.

explainers/compounding-disparities.md - How Do Small Disparities Compound Across Stages?

Selection rates multiply, so a ratio of 0.9 at each of four stages is 0.66 overall. Show the arithmetic on a real audit with simulated stages, compare the multiplied ratio with the measured end-to-end ratio, and say when the independence assumption behind the multiplication fails.

Key citation to verify: Cite a primary source on cumulative disadvantage in selection, for example Bohnet, van Geen, Bazerman (2016), "When Performance Trumps Gender Bias: Joint vs. Separate Evaluation" (Management Science) - confirm.

explainers/differential-validity.md - What Is Differential Validity?

Whether a score predicts the outcome equally well for each group: compare the correlation or AUC between a test score and the outcome per group, with intervals, on a real audit, and relate it to predictive parity and to aggregation bias. Cover the difference between differential validity and differential prediction (different intercepts and slopes).

Key citation to verify: Cleary (1968), "Test Bias: Prediction of Grades of Negro and White Students in Integrated Colleges" (Journal of Educational Measurement) - confirm.

Related explainers: disparate-impact, predictive-parity, aggregation-bias, multiple-comparisons, bootstrap-confidence-intervals, intersectional-bias.

See CONTRIBUTING.md: add explainers/<slug>.md plus a one-line entry in assets/explainers-data.json, then run make build-explainers; the build script generates the pages, sitemap, llms-full.txt and OG images, and llms.txt and the README table must list each explainer (scripts/check_explainer_count.py checks both). Match the structure of a close sibling explainer, use real code run against one of the repo's audits (faircode.benchmark.load_audit loads an audit exactly like the benchmark harness) and real, not invented, results, and if you quote a Fair Code result say whether it comes from results/ (current) or paper/results-frozen/ (the earlier reference snapshot), per CLAUDE.md. Verify every citation yourself before using it.

Done when: the 3 explainers, their README, CONTRIBUTING, ROADMAP, llms.txt and CHANGELOG entries, the website JSON entries and the regenerated generated files are merged, make lint passes, and each explainer's code is runnable from the repo root.