Results that did not hold, with the verdict each one carries in the ledger.
A campaign that only shows its wins is showing you a filter, not a record. These are the results this ship registered, flew, and then could not certify — each with the verdict label written in the findings ledger, copied here exactly rather than restated. Several of them cost more to establish than the wins beside them, and two were killed by the ship's own red team before anyone outside saw them.
Verdict labels are the ledger's own words, not a summary of them.
Finding 10's "noise narrows confidence intervals 34–63%" is KILLED: at matched oracle budget the narrower noisy interval has 0% coverage (vs 95% noiseless) — the estimator lands tightly around the wrong answer.
Grover amplification of the QQQ tail survives on hardware through k=4 (refuting F54's own "garbage by k≈5" pessimism — the contrast peaks at k=4) — but the honest blind multi-k estimate is ~12× worse than just reading the shallow loader: no practical QAE win.
Honest self-correction: a proposed "independent" DAG-fit corroboration turned out to be an exact rescaling of the witness itself (residual = 2.25·DISC, R²=1.0 to machine precision) — a tautology, retracted before being run.
The final invoice — Landauer floor of the ICO-engine demon record (Horizons-3 H4, an honest STRADDLE-REFUTED loss): the engine qubit's Landauer erasure floor [0.1234, 0.1539] E (from measured effective temperature) vs the banked F95 work credit 0.0920 ± 0.0098 E — the demon appears to pay its bill …
⊘ SUPERSEDED as-executed / QUALIFIED in principle (own red-team, Ember C4215, pre-submission — audit).
🏆 WIN — certified runtime advantage, fully fenced, 3-of-3 graded: sealed t=80 (80-T, 10-CCZ) Roetteler hidden-shift string recovered blind and EXACTLY (ŝ==s, HD-0) at d2q=167 on kingston, behind: clean-ladder pre-gate (both free t=0 rungs exact, register certified BEFORE the seal opened), …
THE QUORUM FACT — registered threshold-shape NOT ESTABLISHED (FAIL → DOES NOT HOLD → UNDERPOWERED across three progressively refined flights), and two mechanisms CONFIRMED en route.
THE TIME FLIP — a process in coherent superposition of FORWARD and BACKWARD time direction wins the discrimination game at 0.9953–0.9984, beating the definite-time-direction ceiling at 113–200σ, replicated across three flights on two backends — and the verdict is still DOES NOT HOLD.
HEAT FLOWING BACKWARD — NOT HELD; the correlations bought total SUPPRESSION of a 22σ thermal flow, not its reversal.
THE WINDING METER, S0 — NO-FLY AT THE GATE, and the pilot priced a standing ceiling.
THE VACUUM MINE, third shaft — harvesting DOES NOT HOLD, and the null calibrated a standing ceiling.