Playbook

1 · This quarter

• Inventory where personality scores decide outcomes. List every stage where a personality or integrity-adjacent score ranks candidates, sets a cutoff, or feeds a composite. The costs in the calculator apply where scores drive top-down decisions; advisory-only uses carry less exposure.
• Ask your vendor one question: which samples produced your norms and intercorrelations? If the answer is incumbents, students, or "research samples," the numbers describe people with nothing at stake. Ask for applicant-sample evidence for your region and instrument. The correlation between Conscientiousness and ability differs by setting — and the difference is the whole story here.
• Turn on the cheap countermeasures. Verification warnings (telling applicants responses may be checked) and forced-choice response formats are the two the manuscript names. The case for them is prevention: by the time faking shows up in outcomes, no aggregate statistic you monitor will have warned you.
• If integrity matters for the role, measure it directly. Inferring integrity from a Conscientiousness composite uses exactly the channel faking corrupts. In every model specification that fit the published targets, the applicants who gained seats through faking were 0.2–0.8 SD lower in integrity than those they displaced.
• Re-read your last bias audit knowing what it cannot see. Selection-rate impact ratios stay essentially undisturbed under faking in these simulations. A passed audit says nothing about whether a fifth to a third of your hires are different people than honest answers would have produced.

2 · Structuring the selection system

• Price the stakes into every validity number. The core design principle from Study 1: predictor intercorrelations, and therefore composite validity and adverse-impact projections, are setting-specific. When you model a composite, use applicant-context inputs for your region — not the convenient research-sample matrix.
• Let ability and verifiable evidence carry the ranking weight. Cognitive measures are hard to fake upward; self-report personality under high stakes is demonstrably not. Where personality enters a ranked composite, know that its effective weight includes an inflation component that is largest for your weakest applicants.
• Position self-report personality where stakes on it are lower. The distortion documented here is a high-stakes phenomenon: the same instruments behave normally in research administrations. Screening-out on extreme scores, post-offer development profiling, and structured-interview seeding are uses that keep the information while shrinking the incentive to fake the ranking.
• Build the counterfactual check into monitoring. The damage lives in who gets hired, so monitor composition, not just rates: track hire-cohort performance against score bands over time, and treat a weakening personality–performance gradient among hires as a faking signal, not noise.
• Validate locally, in the applicant context. Everything above is a projection from published evidence through an explicit model. The projection beats assuming zero cost, but a local criterion study on genuine applicants beats the projection.