Iteration 58 — the Cousins Split H-060 M3 CLEAR H-061 EXCLUDED (η 0.897, KNIFE EDGE)

In plain words: the field test split the pair. The worst-regime gap meter — the cousin with no dials at all — came through clean: readings spread across the whole range, and its strongest resemblance to anything certified is 63%. The tail-risk cousin fell on the narrowest margin of the campaign: its rankings at two adjacent tail sizes agree 89.7% — three thousandths below our 90% bar. Full honesty on the record: the instability lives only at the least tail-like end of its swept range (the deep-tail end was rock-stable at 97.7%), and a friendlier choice of range would have acquitted it. But redrawing a pre-registered range AFTER seeing the results is precisely the move this campaign forbids — the same discipline that killed an earlier candidate at its range edge must apply symmetrically here, or pre-registration means nothing. The kill costs little (the surviving cousin covers the job, and their readings overlap 77%) and the repair is cheap: derive the tail size from first principles, or pre-register the narrower stable range — a fresh candidate either way. The survivor marches to the final hurdles.