Short answer. An LL144 bias audit is not just an impact-ratio calculation. We first define the AEDT and deployment context, validate the data and output meaning, determine the correct calculation path, perform the required demographic analysis, review edge cases and limitations, and then produce a technical audit record plus publication-ready results.
What we need before calculation
| Input | Why it matters |
|---|---|
| AEDT and use-case description | Defines the tool, workflow and deployment context being audited. |
| Output semantics | Establishes whether the controlled calculation path is selection/classification, scoring or requires further review. |
| Audit population and outcome fields | Defines denominators, target outcomes and the period covered by the audit. |
| Sex and race/ethnicity fields | Supports the required category and intersectional analysis without demographic inference. |
| Job or grouping structure | Helps determine whether results must be analyzed across distinct roles, groups or deployment contexts. |
| Historical/test-data basis | Documents the source of the dataset and whether the data basis is appropriate for the engagement. |
Methodology principle
Code does not become law because it exists, and a professional judgment does not become a DCWP requirement because an auditor prefers it. Our methodology classifies controls so each conclusion can be traced to its proper authority.
Methodology classification
- LL144 / DCWP required — obligations prescribed by statute or rule.
- LL144 / DCWP permitted — choices expressly permitted by the rule, such as the less-than-2% category exclusion.
- Professional auditor methodology — judgments necessary to perform a defensible audit where the rule does not prescribe every statistical decision.
- Lexara internal control — evidence, provenance and quality controls used to make the work reproducible.
Score-based AEDT analysis
For supported score-based AEDTs, AUDITLL144 applies the scoring-rate and impact-ratio framework required by 6 RCNY §5-301, including the rule's median-based analysis. The public methodology describes the regulatory framework and assurance principles; implementation details, normalization logic, validation controls and exception handling remain proprietary.
Selection and classification analysis
For supported selection and classification outputs, AUDITLL144 applies the selection-rate and impact-ratio framework required by 6 RCNY §5-301. Population construction, source normalization, internal mappings and calculation-control logic are maintained within Lexara's controlled audit environment and are not published.
Comparator and denominator controls
Comparator and denominator decisions are reviewed against the applicable rule, the defined audit population and the documented AEDT use case. The public site does not disclose internal implementation logic, tie-handling routines, mapping rules or exception controls.
Demographic controls
Lexara uses supplied demographic information and explicit mapping controls. We do not infer or impute protected characteristics. Known but non-analytical values are preserved as controlled states rather than silently reassigned.
The less-than-2% rule
DCWP permits an independent auditor to exclude a category that represents less than 2% of the data used for the audit from required impact-ratio calculations. Lexara calculates the relevant share but does not make the exclusion automatically; it remains an explicit professional auditor decision.
Quality assurance and professional review
The calculation layer is subject to professional review and documented quality controls. Before reporting, we verify the audit population, data integrity, required demographic analyses, applicable calculations, material limitations and consistency between the technical report and the publication-ready Summary of Results. Detailed QA rules and internal control thresholds are proprietary.
What the client receives
- Technical Audit Report — scope, data basis, methodology, calculations, limitations and review record.
- Impact-ratio and rate tables — for the supported calculation path and required demographic categories.
- Publication-ready Summary of Results — structured for the information required to be made publicly available, without publishing the full confidential audit record.
- Scope and limitation statement — identifying the AEDT, deployment context, data basis and relevant constraints.
View the synthetic Technical Audit Report · View the synthetic Summary of Results
Technical edge cases
- Unknown and missing demographic data
- The LL144 2% rule
- Small samples and statistical significance
- Job groups and multiple positions
What the methodology does not claim
The calculation layer does not automatically issue legal or discrimination conclusions. We do not label an AEDT “bias free,” “compliant,” “non-compliant,” “legal,” “illegal,” “pass” or “fail” solely from a calculated impact ratio.
Evidence, reproducibility and proprietary controls
The Lexara audit platform maintains a controlled path from source evidence through professional review and reporting so reported metrics can be supported and reviewed. AUDITLL144 publishes enough methodology to explain the regulatory basis, audit approach and limitations, but does not publish proprietary software, internal methodology version identifiers, preprocessing rules, mappings, exception logic, QA thresholds, test suites or operational workflows.
This boundary is intentional: clients receive a documented and reviewable audit methodology without disclosure of Lexara's confidential implementation know-how.
Methodology FAQ
No. The public methodology explains the regulatory framework, audit approach, evidence controls and limitations without publishing proprietary preprocessing, mappings, exception logic, QA thresholds or source code.
No. The calculation layer produces the rates and ratios required for the supported audit path; it does not automatically issue a legal, discrimination, bias-free or compliance verdict.
No. AUDITLL144 uses supplied demographic information and controlled mapping states rather than guessing protected characteristics from names, geography, photographs, resumes or similar proxies.