AEDT Bias Audit

Compute the bias-audit statistics New York City Local Law 144 requires before an Automated Employment Decision Tool (AEDT) may be used to screen candidates or employees. The law (NYC Admin. Code §20-870 et seq., in force since 2023-07-05; DCWP rules 6 RCNY §5-300 et seq.) requires an independent annual audit that reports, for each demographic category, the rate at which the tool selects (or scores above the median for) that group and the impact ratio — that group's rate divided by the rate of the most-selected group. An impact ratio below the four-fifths (0.8) threshold from the EEOC Uniform Guidelines flags potential adverse impact; the law requires the number to be calculated and published, not any particular remediation.

The audit is computed across three dimensions: sex, race/ethnicity, and the intersection of the two, using the EEOC categories. Categories that make up less than 2% of the audited data may be excluded from the impact- ratio calculation at the auditor's discretion (DCWP §5-301). This primitive takes the per-category counts — selected/total for a pass-fail tool, or scored-above-the-overall-median/total for a continuous-score tool — and returns the selection (or scoring) rate, impact ratio, and adverse-impact flag per group, plus the most-selected group and an overall flag. It is a pure calculation: the operator supplies the data an independent auditor collected, and gets back the figures the published summary must contain.

b.ai.aedtBiasAudit(opts) #

{
  type:             string,   // "selection" | "scoring" (required)
  metadata:         object,   // { tool, auditor, auditDate, distributionDate? } (tool/auditor/auditDate required)
  categories:       object,   // { sex?, raceEthnicity?, intersectional? } → { : { selected|scoredAboveMedian, total } }
  minCategoryShare: number,   // default: 0.02 (DCWP §5-301 — sub-2% categories may be excluded)
}

Compute the NYC Local Law 144 bias-audit figures from per-category counts. type is "selection" for a pass-fail tool (each group entry is { selected, total }) or "scoring" for a continuous-score tool ({ scoredAboveMedian, total }, where the count is candidates scoring above the overall median). Returns the selection/scoring rate, impact ratio (group rate ÷ most-selected group's rate), and an adverseImpact flag (impact ratio < 0.8) per group, across the sex, raceEthnicity, and intersectional dimensions, plus the most-selected group per dimension. Categories under minCategoryShare (2% by default) are marked excluded and left out of the impact-ratio basis. Throws AedtBiasAuditError on malformed input. The result is the data an employer must publish; the law mandates the calculation, not any remediation.

var report = b.ai.aedtBiasAudit({
  type: "selection",
  metadata: { tool: "ResumeRanker v3", auditor: "Acme Audit LLC", auditDate: "2026-05-26" },
  categories: { sex: { Male: { selected: 60, total: 100 }, Female: { selected: 42, total: 100 } } },
});
report.results.sex[1].impactRatio;   // → 0.7  (Female: 42% / 60%)
report.results.sex[1].adverseImpact; // → true (below the 0.8 four-fifths threshold)

Last updated 2026-08-08T16:39:49.652Z by seeder.