Iteration 65 — H-062 W-DRO PARKED with a PROPOSAL CERTIFICATE VOID ON DEPENDENT DATA · NO COMPUTE SPENT

In plain words: the next candidate's entire selling point is a NUMBER with a guarantee: "your conclusion survives this much distribution shift, with this confidence." But the guarantee's fine print assumes independent samples — and market bars are anything but independent: on our data the guaranteed radius could be wrong by a large factor, which makes the number exactly the kind of confident-but-unbacked reading this campaign exists to keep off the shelf. Examining the guarantee-less version instead would quietly turn the candidate into yet another loss-spread readout — the family we just executed three of — a foregone kill that tests nothing the candidate actually claims. So the honest move is neither: the candidate is PARKED, with a written proposal spelling out exactly what a valid exam requires (a dependence-aware recalibration of the radius with a derived block length, plus a coverage check against our eight real regime shifts), and a re-harvest lead filed for dependence-robust variants of the method that exist in the literature. Zero compute was spent on a question that arithmetic already answers. The batch's final candidate is next.