Do the search's finds still pass when re-checked on the corrected ladder?

EXPERIMENTS.md · lines 16186–16189

E240 — E223's finds re-walked on the corrected ladder (E236 redone; pre-registered 2026-09-23 17:2x)

E236's design, now runnable: all 32 walked finds, walk_finds.py --top lattice, E223's environment, rung-1 grid 2600 → 100 K plus 80 and 60 K, the floor rule, and rung 0 AND rung 1 on

EXPERIMENTS.md · lines 16271–16286

E240 RESULT (22:0x) — and two defects in rung 1's floor handling, caught before any DFT. All 32 finds walked (E237's e6 ensemble, exact evaluator, grid to 60 K). Checking the E241 survivor's curve before spending A100 time exposed (i) the floor rule read only the energy-slope channel: five finds whose slope maximum sat at the 60 K floor had the FLUCTUATION channel peaking inside the window (331, 353, 364, 380, 575 K) — in equilibrium both channels are the heat capacity, so a slope maximum at the floor is relaxation, and those five order; and (ii) the lattice rung scored every floor pass as "censored" while the headline p kept rung 1's floor value, so rung 2's decomposition drive never reached them (Mo₇₉Hf₁₉, Mo₇₀Hf₂₀Ti₁₀ kept p 0.78 with MACE drives of +61 and +35 meV/atom). Both fixed in forager/ladder/verifier.py with tests; E240's saved results re-scored through the fixed code without re-sampling (scripts/validate/e240_rescore.py → runs/e240_rescored/): 18 of 32 p values change; 11 finds pass (p 0.50–0.77, both channels silent to 60 K, rung 2 stable). The original survivor pick (Mo₅₂W₂₅Ta₂₂) was one of the five and is void. (1) CONFIRMED — 24/24 of E223's no-verdict finds carry a verdict. (2) FALSIFIED — the Mo–Ta-rich finds do not order higher under e6: Mo₄₂Ta₂₉Ti₂₉ 896 → 795 K, Mo₄₈W₂₉Ta₁₂Nb₈ 424 → 211 K, Mo₅₄Ta₁₆W₁₃Ti₇Hf₅ 569 → 626 K, although e6 binds B2 Mo–Ta ~35 meV/atom deeper — so e6's rung 1 gets a known-answer check (E242 (4)). (3) CONFIRMED — the top-p pick changed (Mo₄₂Ta₂₉Ti₂₉ 0.57 → Mo₅₉Ti₃₀W₁₁ 0.77).