Four-Fifths of What? Statistical Indeterminacy in NYC Local Law 144 Bias Audits and the Governance Parameters It Exposes

New York City's Local Law 144 requires employers using automated hiring tools to publish bias audits that report impact ratios, but sets no threshold, error tolerance or decision rule. Reanalysing every public LL144 audit in the ACLU Auditing the Audits corpus (114 audits, 2,154 impact ratios), this preprint finds that 22.1% of reported ratios cannot be distinguished from the four-fifths benchmark by the data behind them, and that with familywise error control only 29 of 114 audits support a claim that some ratio is below 0.80 and 27 that all are at or above it. It then shows that one parameter the city's rule leaves to the auditor decides outcomes on its own. The discretionary exclusion of categories under 2% of applicants (6 RCNY 5-301(d)) decides the four-fifths outcome of 22 of 114 audits (14 employers, 8 audit firms); under the point reading the option can only favour the audited employer; 19 of the 22 published the passing version; and 28 of the 34 groups it removes from the reference position are American Indian, Alaska Native, Native Hawaiian or Pacific Islander. The dependence holds under interval readings (21–30 audits), when tiered tools are judged on one selection definition (20–35), for 17 of the 22 under every count consistent with the published rounded rates, and without the largest audit firm (14–16 of 44 audits). The paper proposes a three-state reporting standard with parameter provenance, and community mandates: machine-readable, content-hashed rulebooks through which a public body, rather than the audited party's contractor, sets the parameters that decide whether an audit passes. A replication archive with all code, the mandate schema and evaluator, and generated numbers is included; 33 automated tests reproduce the reported figures exactly. Not peer reviewed. No wrongdoing by any named employer or auditor is alleged.