Fairness Certificate
Issued April 04, 2026 — Paragon v1.0
Color Health
This certifies that the AI model identified below was trained with
algorithmic fairness constraints using the Paragon bottleneck method.
R²
0.2217
Task Performance
Equity Ratio
0.3750
worst / best group
Bottleneck d
32
latent dimension
Dataset
2,504
samples · 5 groups
Sensitive Attribute Leakage
Equity Gap: 26.8ppMODERATE
Per-Group R²
Mean: 0.0974 | Range: 0.0480 – 0.1810 | Spread: 0.1330
Regulatory Compliance
EU AI Act Article 10
Data & Data Governance
PARTIAL
EU AI Act Article 15
Accuracy & Robustness
PARTIAL
Automated assessment. Full compliance requires legal review.
Model Details
| Model ID | a1b2c3d4e5f6789012345678901234567890abcdef1234567890abcdef123456 |
| Sensitive Attribute | ancestry |
| Target Attribute | height_prs |
| Leakage | 46.8% |
| Chance Baseline | 20.0% |
| Equity Gap | 26.8pp |
| Issued (UTC) | 2026-04-04 12:00 UTC |