Calibration CheckLive
Is your AI's 90% really 90%?
Give it labeled examples and a model's probabilities. See whether its confidence can be trusted, fix it with one number, and pick the cut-off that costs least. Runs in your browser; your file never leaves the page.
- Calibration error, Brier score, log loss and AUC, against a baseline
- Two fixes compared on held-out cases: temperature and isotonic
- The cost of every cut-off, and the most confident mistakes