PD Gap Detection
The gap detection module identifies "blind spots" — regions of the design space where model coverage is insufficient.
- The final validity scores are converted to a binary coverage map:
$v_{i,k}\ge0.2$ → covered. - A decision tree classifier is trained to partition the design space into regions of high vs. low coverage.
- The tree is traversed to extract interpretable rules describing each blind spot region.
- Coverage statistics are reported: what fraction of the design space is covered at each validity threshold.
Gap detection routine
