Verification · Finding validation · Confidence calibration
Calibration curve
Fit confidence against accepted or rejected outcomes.
How to build it
- Log each posted finding with its reported confidence and the human outcome.
- Fit a calibration curve from the logged pairs on a monthly schedule.
- Save the fitted curve and the data it was fitted on with the report.
Signals in a repository
- A script that imports calibration_curve from scikit-learn.
- A log or table of finding outcomes with confidence values.
- A scheduled workflow that refits the curve and stores it.
In practice
No scanned repository runs a tool for this check. The steps above are the only guidance.