Fairness Blueprint
EEOC / UGESP Adverse Impact Compliance
Commitment to Fair Assessment
Cognaium assessments are designed to minimize disparate impact across all protected groups. We follow the Uniform Guidelines on Employee Selection Procedures (UGESP, 29 CFR Part 1607) and apply the four-fifths rule as a monitoring threshold.
Our goal is to ensure that AI-assisted assessments are job-relevant, validated, and fair to all candidates regardless of race, gender, age, disability, or other protected characteristics.
Adverse Impact Methodology
The Four-Fifths (80%) Rule:
The selection rate of any demographic group should not be less than 80% of the selection rate of the highest-selected group. When the impact ratio falls below 0.80, this indicates potential adverse impact requiring further investigation.
How we calculate:
- Calculate the selection (pass) rate for each demographic group at each assessment stage
- Identify the group with the highest selection rate
- Divide each group's selection rate by the highest rate to get the impact ratio
- Flag any impact ratio below 0.80 for investigation and remediation
Protected Categories Monitored
| Category | Groups | Legal Framework |
|---|---|---|
| Race / Ethnicity | EEO-1 categories (Hispanic/Latino, White, Black/African American, Native Hawaiian/Pacific Islander, Asian, American Indian/Alaska Native, Two or More Races) | Title VII, UGESP |
| Gender | Male, Female, Non-binary/Other | Title VII, UGESP |
| Age | 40+ vs. Under 40 (per ADEA threshold) | ADEA |
| Disability Status | Disability disclosed vs. not disclosed | ADA |
Adverse Impact Analysis Results
Pending First Audit Cycle
Results from our most recent adverse impact analysis will be published here following completion of the first formal audit cycle. Our internal bias audit schedule requires quarterly internal audits and annual independent audits.
For NYC Local Law 144 bias audit data, see our Bias Audit page.
Model Change Update Policy
This analysis is updated:
- With each material model change (provider swap, prompt version, scoring algorithm change)
- At minimum quarterly as part of our internal bias review cycle
- Annually through independent third-party audit
Date of last analysis: [Pending first audit cycle]
Fairness Controls in Place
- LanguageBiasGuard: Adjusts scoring for ESL (English as a Second Language) patterns to reduce language-based disadvantage
- FairnessMonitor: Real-time monitoring of score disparity across demographic groups with alerting when thresholds are exceeded
- FairnessEvaluationPipeline: Daily aggregation of scoring data by subgroup with automated anomaly detection
- Human Review Requirement: All adverse hiring decisions require human confirmation before action
- Quarterly Bias Audits: Internal review of selection rates and impact ratios across all protected categories
- Job-Relevance Validation: All assessment content is generated from and validated against specific job requirements
Questions About Our Fairness Practices
For questions about assessment fairness, adverse impact methodology, or to request detailed fairness data, contact [email protected].