Verifies that a degradation curve is monotone non-increasing: injecting more failures must never improve the Submission Confidence Index. A rising step would signal a defect in the scoring model.
Arguments
- curve
A tibble from
sci_degradation_curve(), or any data.frame withstepandSCIcolumns.- tol
Numeric tolerance for treating a small rise as noise. Defaults to
1e-8.
Value
A list with:
monotone:TRUEif no step rises beyondtolmax_increase: the largest step-to-step increase (0 if none)violations: a tibble of the steps that rose, empty if none
Examples
if (FALSE) { # \dontrun{
curve <- sci_degradation_curve(evidence)
sci_monotone_check(curve)$monotone
} # }