Iteration 89 — H-046 Romano–Wolf Passes the Exam EXAM PASS · FWER 0/49 WHERE HAC HIT 43% · THE p-SOURCE FOUND

In plain words: yesterday's disease met today's cure. The previous iteration discovered that the standard time-series p-values feeding the declaration layer are broken on this data — a 5% error budget produced false discoveries 43% of the time. This iteration put the stepdown procedure's OWN inference on trial with the identical torture test: forty-nine worlds where every discovery is false by construction. The result: zero false discoveries in all forty-nine worlds. The difference is philosophical as much as technical — instead of trusting a textbook formula for how much uncertainty persistent data carries, the stepdown measures its uncertainty empirically, from time-rotations of the data itself, so the dependence is baked into the null rather than approximated. On the real panel it still finds fifteen genuine discoveries (the broken pipeline had claimed sixty-one). One honest design ruling is on the record: the textbook version of this procedure uses sign-flipped residuals, which sit in a gray zone of the no-fabricated-data amendment — the clean rotation-based version was used, and the gray zone is flagged for the operator to rule on, not decided unilaterally. The declaration layer has its foundation candidate; the false-discovery backbone killed last iteration re-enters behind it if the ladder completes.