Fairness testing and evaluations
Check whether a model treats groups differently, and score it on robustness, leaks and hallucinations.
Fairness testing
Article 10 asks providers of high-risk systems to examine data for possible bias. Upload a CSV of your model's predictions with the group column you want to check (for example age band or gender), or try a sample file.
V77 Trust compares outcomes across groups using standard measures: the selection-rate ratio between groups (demographic parity), and differences in true- and false-positive rates (equalised odds and equal opportunity). It flags differences above common thresholds.
A flagged difference is a signal to investigate, not a legal finding of discrimination. Record what you found and what you did about it.
Evaluations
Evaluations score a model on robustness, prompt injection, personal-data leaks and hallucinations, so you can compare versions and track quality over time.
Using the results
- Attach reports as evidence on the accuracy, robustness and data-governance controls.
- Feed results into impact assessments; V77 Trust can draft them from your test results.
- Re-test when the model, data or use changes.