Iteration 84 — H-072 MRP EXCLUDED · B-04 CLOSES KILL #25 · BATCH CLOSED 12/12: 2 ADMITS · 10 KILLS

In plain words: the batch's last candidate won the fight everyone was watching and lost to a technicality that isn't one. Against the certified flip counter — the fight its dossier flagged from birth — its worst-regime floor reading proved genuinely different (9% overlap, a decisive win). But when the same reading was recomputed on a slightly smaller data sample, the feature ranking reshuffled wholesale (74% agreement, far under the 90% bar). The reason is structural: a MINIMUM is set by whichever single regime happens to score lowest, and with noisy estimates, different samples crown different worst-regimes — the same disease that killed the tail-variance candidate, in minimum's clothing. The boneyard entry says out loud: a noise-controlled floor (a soft minimum) would be genuinely worth building, because the orthogonality is real. With that, batch B-04 closes twelve-for-twelve: TWO certificates (both distance geometries — the boundary that rested on one instrument now has three) and TEN kills, each by a different named mechanism — a taxonomy of how instruments fail, built one honest execution at a time.