Comparing the AUT results against the actual "true" size of the seeded flaws (often verified later by macro-sectioning the weld).
DNV-RP-F118 is known for its strict statistical thresholds to ensure high confidence in the data: PoD Criterion: dnv-rp-f118