Iteration 67 — B-04 Opens: MMD Passes Exam v2 Bit-Exactly NEW BATCH · EXAM PASS (v2) · M3 NEXT

In plain words: with the STABLE batches done, the queue turned to the boundary in worst shape: the FRAGILE-vs-CONDITIONAL split currently rests on a SINGLE certified instrument (the regime counter) — a single point of failure the new batch of twelve exists to fix. Its cheapest candidate — a kernel distance that asks "does this feature's value distribution differ between two regimes?" with no classifier and a built-in shuffle test — passed its exam bit-exactly, after one instructive stumble: the first exam form demanded the distance between a series and its byte-twin read zero, but the textbook's UNBIASED estimator deliberately reads slightly negative on identical samples (a fact about the estimator, not a flaw) — the exam was wrong, the campaign's own fix-the-exam-on-the-record precedent applied, and form v2's correct criterion (the twin must match the series' own self-distance to the last bit) passed with differences of exactly zero everywhere, while real regime differences rang the bell at up to fifty times the shuffle bar. Field test next, with the incumbent regime counter as the named comparison.