Iteration 86 — H-093 GW-CPA EXCLUDED KILL #26 · DIAL LAW 0.868 — THE CLOSEST MISS · EXAM-INTEL FOR H-092

In plain words: the declaration layer's opening candidate died the campaign's narrowest death. Recomputed on a slightly smaller data sample, its feature ranking agreed 86.8% with itself — the bar is 90%, and the rule fires mechanically: an instrument whose rankings reshuffle a tenth under an innocuous data-volume change certifies nothing. What makes this kill different from the floor statistic's (73.5%) is everything else: against the entire certified stack its worst overlap was 37% — the cleanest sheet any candidate has shown — and its named-adjacency fights (the calibration meter, the flip counter) were won outright. Two durable goods fell out. First, intelligence: the regime-conditioned reading genuinely differs from the plain unconditional test (74% related, well under echo territory) — so the next candidate in line, the classic Diebold–Mariano, is NOT pre-killed and enters on its own merits. Second, a precise successor prescription on the boneyard: a variance-stabilized version of this exact instrument is the batch's most promising revisit — the information is demonstrably there; only the estimator wobbles.