Short answer. A defensible LL144 audit starts before any calculation. The auditor needs a clear AEDT use case, a documented data basis, explicit output meaning and enough evidence to reproduce the population and calculations. AUDITLL144 separates those steps rather than treating a CSV upload as the entire audit.
1. Engagement triage and independence review
We identify the AEDT, the hiring or promotion workflow, the output type, the relevant organization/vendor relationships and whether the proposed engagement presents an independence issue under 6 RCNY §5-300.
The audit software does not automatically decide legal scope or auditor independence. Those are professional review questions.
2. Data-source and evidence review
DCWP treats historical data as the ordinary basis for a bias audit and permits test data in specified circumstances when sufficient historical data is unavailable. The summary of results must explain the source and data used.
We document the source dataset, output meaning, relevant population, demographic fields and the basis for any use of test data.
3. Demographic mapping without inference
DCWP's FAQ states that imputed or inferred demographic data cannot be used to conduct the audit. Lexara therefore does not infer sex or race/ethnicity from names, photographs, geography, language, resumes or similar proxies.
Source values are preserved and mapped only through explicit controlled rules. Unknown, declined, missing and outside-taxonomy states are not silently converted into an analytical category.
4. Determine the supported calculation path
For AEDTs that select candidates to move forward or classify candidates into groups, the rules require selection rates and impact ratios. For AEDTs that score candidates, the rules require the full-sample median, scoring rates and impact ratios. AUDITLL144 maintains separate implemented methods for these output families.
5. Required demographic analyses
Applicable calculations are performed separately across sex categories, race/ethnicity categories and intersectional sex × race/ethnicity categories. For classification outputs, required analysis is performed for each specified classification group where the controlled methodology supports the output.
6. Less-than-2% review
DCWP permits an independent auditor to exclude a category representing less than 2% of the data from required impact-ratio calculations. Exclusion is not automatic. When invoked, required count/rate information and the auditor's justification remain reportable.
7. Professional calculation review
Deterministic calculations are reviewed against the exact evidence and methodology state. The platform is designed not to convert an impact ratio into a legal conclusion such as “pass,” “fail,” “bias free” or “discriminatory.”
8. Reporting
Audit outputs are structured to support both technical review and the information needed for the public summary under 6 RCNY §5-303. The employer or employment agency remains responsible for making the required information publicly available before covered use.
What to send first
- AEDT/vendor name and version, if known.
- Whether the tool is used for hiring or promotion.
- What output the tool produces: score, selection, classification, ranking or other output.
- Whether historical use data exists.
- Approximate audit deadline.
Do not send candidate-level demographic data through ordinary email.
Last legally reviewed: September 17, 2026.