Iteration 62 — H-064 Calibration Gap Passes the Entrance Exam EXAM PASS · IPP-SHADOW CHECK DECISIVE NEXT

In plain words: the second-to-last candidate of the batch reads a genuinely different quantity from the fallen loss-spread family: not how BIG a model's errors are per regime, but whether the model's confidence statements stay honest in each regime — a published theorem ties "calibrated in every regime" to the very invariance our certificates are about. Its exam was the cleanest of the diagnostics so far: the perfect-by-construction model reads zero divergence (through the pre-registered handling of its peculiar edge case), and the deliberately wrong model's calibration divergence beats everything label-shuffling of the same real values can produce by a factor of six — at every choice of the binning knob, which was swept on schedule. But its decisive fight is named and it's a familiar one: our certified IPP instrument already reads distributional honesty per feature — if the calibration lens turns out to be IPP's shadow, the redundancy gate ends it. The field test decides.