Iteration 88 — H-047 BY EXCLUDED by Input Invalidity KILL #28 · THE DISCOVERY OUTRANKS THE KILL · OPERATOR FLAG

In plain words: the false-discovery-control backbone died at the door — but not by its own fault, and the autopsy produced the iteration's real headline. The procedure itself was verified exactly correct: every algebraic identity held, and the byte-twin's significance matched its original's to the last bit. Then the control claim went on trial with the harshest possible design: targets circularly rotated inside each regime so that every feature-target relationship is destroyed by construction — every discovery in that world is false. A 5% error budget should mean false alarms in about one replicate in twenty. The result: 43% of replicates produced false discoveries (uncorrected: 62%). The culprit is upstream: the standard time-series p-values feeding the procedure understate uncertainty on data this persistent — and no correction procedure can launder invalid inputs. The kill has a built-in resurrection clause: BY re-enters automatically once a valid p-value source is certified — and the NEXT candidate in the queue, Romano–Wolf, is precisely a resampling-based inference that might be that source. A proportionate flag goes to the operator: certified instruments that used these p-values sat behind a far stricter threshold (99.9%) — margin, not immunity.