LLumen

Ethical Design

Bias Explorer

Swap training distributions and watch fairness and accuracy metrics move.

Bias Explorer

Swap training distributions. Overall accuracy can look fine while one group is left behind.

Equal examples from Group A and Group B; careful labels.

Overall accuracy

87%

Group A accuracy

88%

Group B accuracy

86%

Fairness gap: 2 points. A serious team reports disaggregated metrics — not averages alone.