Iteration 91 — the SPA–MCS Pair Splits H-095 MCS PASSES · H-094 SPA KILLED (#30) · k_v COUNT FIGHT NEXT

In plain words: two set-minded candidates entered together and split. The Model Confidence Set — which answers "in WHICH regimes is a feature's advantage indistinguishable from its best?" with a statistically guaranteed set — passed all three doors: the byte-twin identity held; its coverage guarantee was verified on twenty-five worlds where all regimes are equal by construction (it wrongly shrank the set in only two, inside the allowance); and on the real canonical pair it RESOLVED structure, scoping out exactly one of eight regimes as genuinely inferior. That is set-valued CONDITIONAL scoping working at the door — the first candidate of the batch to demonstrate the boundary's native question. Its sibling — the snooping-corrected "is the best regime really superior?" test — missed its own bar (p = 0.07 against 0.05) on the very pair where regime structure is certified at fifteen-million-to-one, and died by the blindness rule; the miss is marginal and the boneyard says so, but MCS covers the scoping need, so the revisit is low-priority. Next comes the fight that will decide MCS's fate: its per-feature reading is a COUNT of regimes — and the certified incumbent k_v is also a count of regimes. Same grammar, possibly same information: the census will say.