Iteration 93 — H-090 Exceedance Correlation EXCLUDED KILL #32 · REAL ASYMMETRY, REGIME-INVARIANT · EXAM KILL

In plain words: the characterization layer's first candidate died the most philosophical death of the campaign: it measured something TRUE that lives at the wrong boundary. The instrument reads how much more tightly a pair co-moves in joint crashes than in joint rallies. First, an honest stumble: fed the campaign's canonical pair — which moves in OPPOSITE directions by nature — the first form crashed because the "both high" corner of the data is nearly empty; version two orients the pair by its overall association sign (derived, no knob) and ran clean. Then the discovery: the crash-versus-rally asymmetry is REAL and cleanly measurable — about 0.25 extra correlation in the crash tail, confidently resolved inside each regime, exactly where the earlier local-dependence family drowned in noise. And then the kill: that asymmetry is nearly IDENTICAL in the covid crash and in the calm mid-cycle (difference 0.04 against a noise floor of 0.21). A reading that never changes across regimes carries zero regime-boundary information — ironically, it is STABLE-flavored structure, and the boneyard notes that a per-feature form of it could re-enter someday under a STABLE-boundary claim. The page also carries a small bookkeeping correction: the previous steering said five candidates queued; the correct count was six — five now remain.