Independent Bias Audit Report

Auditor Certification

Executive Dashboard

Executive Summary

This independent bias audit of TalentScreen AI finds the AEDT to be in substantial compliance with NYC Local Law 144 requirements. Statistical analysis of 3,470 applicant records across 3 job categories reveals no adverse impact under the EEOC 4/5ths rule for any protected category. All impact ratios exceed the 0.80 threshold.

Key Findings

• ✓ No adverse impact detected across sex or race/ethnicity categories
• ✓ All impact ratios above 0.80 (4/5ths rule) threshold
• ✓ Chi-Square tests show no statistically significant disparities (p > 0.05)
• ✓ Data quality score: 94% completeness across all demographic fields

Compliance Status

• ✓ Independent bias audit: Complete — certified by AI Compliance Officer
• ✓ Public disclosure page: Published — hosted at haiec.com
• ✓ Candidate notification: Template provided — 10 business day notice
• ⓘ Advisory: Consider implementing intersectional analysis for future audits

Selection Rate Analysis

Selection Rates by Sex

Category	Applicants	Selected	Selection Rate	Impact Ratio	Status
Male	1,892	412	21.8%	1.00 (reference)	PASS
Female	1,578	328	20.8%	0.95	PASS

Selection Rates by Race/Ethnicity

Category	Applicants	Selected	Selection Rate	Impact Ratio	Status
White	1,215	268	22.1%	1.00 (reference)	PASS
Black or African American	486	101	20.8%	0.94	PASS
Hispanic or Latino	625	134	21.4%	0.97	PASS
Asian	798	172	21.6%	0.98	PASS
Native Hawaiian/Pacific Islander	42	9	21.4%	0.97	PASS
American Indian/Alaska Native	28	6	21.4%	0.97	PASS
Two or More Races	276	50	18.1%	0.82	PASS

Note: All impact ratios exceed the 0.80 threshold under the EEOC 4/5ths rule. The "Two or More Races" category (0.82) is above threshold but warrants continued monitoring. Fisher's Exact test applied to categories with fewer than 50 applicants.

Statistical Tests

Chi-Square Test Results

Comparison	Chi-Square Statistic	p-value	Significance
Selection by Sex	0.52	0.471	Not significant (p > 0.05)
Selection by Race/Ethnicity	2.14	0.906	Not significant (p > 0.05)

Fisher's Exact Test Results

Category	p-value	Significance	Reason for Fisher's
Native Hawaiian/Pacific Islander	0.892	Not significant	n < 50
American Indian/Alaska Native	0.934	Not significant	n < 50

Yates' correction applied to Chi-Square tests. Bonferroni correction applied for multiple comparisons. All calculations are deterministic and reproducible (SHA-256 verified).

Methodology

This independent bias audit was conducted in accordance with NYC Local Law 144 and DCWP Rules §5-300 through §5-303, using HAIEC's deterministic analysis engine.

Data Sources

Historical applicant data from January 1, 2025 through December 31, 2025. Data extracted from employer's applicant tracking system via structured CSV export. 3,470 total applicant records across 3 job categories: Software Engineering, Product Management, and Customer Success.

Statistical Methods

• EEOC 4/5ths Rule: Impact ratio calculated as (selection rate of protected group) / (selection rate of reference group). Threshold: 0.80
• Chi-Square Test: With Yates' correction, applied to categories with expected cell frequencies ≥ 5
• Fisher's Exact Test: Applied to categories with small sample sizes (n < 50)
• Bonferroni Correction: Applied for multiple comparisons across race/ethnicity categories
• Reference Group: Highest selection rate group used as reference (per EEOC Uniform Guidelines)

Analysis Engine

All calculations performed by HAIEC Deterministic Analysis Engine v2025.1.0. The engine is rule-based and deterministic — same input data always produces the same output. All results are SHA-256 hashed for tamper detection and reproducibility verification.

LL144 Requirements Assessed

• Annual Bias Audit: Independent audit of AEDT for bias and discriminatory impact
• Public Notice: Publication of audit results on company website (§5-301)
• Candidate Notification: Inform candidates of AEDT use and data collected
• Alternative Process: Provide alternative selection process upon request
• Data Retention: Maintain audit and usage records

Recommendations

Evidence Bundle

The following SHA-256 hashed evidence files accompany this audit report:

File	SHA-256 Hash (first 16 chars)	Description
run-spec.json	a7f3b2c1e4d89f01	Deterministic run specification
input-hashes.json	9c1d4e8fb2a73f06	Input data integrity hashes
analysis-results.json	3e7a9b2d1c8f4e05	Complete statistical analysis output
validation-report.json	b4c8d2e6f1a39507	Data quality and validation report
methodology.md	e2f6a8c4d0b71e93	Methodology documentation
config-manifest.json	7d1e3f5a9b2c8406	Configuration and parameters
disclosure-data.json	5a9c3e7b1d4f8602	Public disclosure page data
verdict.json	c6d8e2f4a0b31975	22-point compliance verdict
bundle-manifest.json	1f3a5c7e9b0d2846	Bundle integrity manifest

HAIEC — AI Compliance Infrastructure | www.haiec.com | compliance@haiec.com