Evidence
Benchmarks and evidence for fairness and stability.
BlindStairs documents how evaluation behavior stays stable across equivalent profiles and controlled variations.
- Paired-profile tests for demographic stability
- Stability checks across time and reviewers
- Leakage detection before results are returned
Bias Variance
Protected Group Stability
Standard
BlindStairs
High Variance
98% Secure
Evidence details
What the evidence covers
Paired-profile benchmarks
Equivalent profiles are evaluated consistently when demographic attributes are swapped or masked.
Stability under perturbations
Outputs remain stable across time, reviewers, and controlled variations in input format.
Leakage detection
Checks for indirect demographic signals that could influence outcomes.
Availability and review
Detailed evidence is available under NDA and included in compliance documentation when required.