IBM HERON2026 CAMPAIGN~110 EXPERIMENTSF01–F121EVERY NUMBER A JOB ID

Have we found quantum advantage?

Yes — measurably, on four of five scoreboards. Four clear exact theorem ceilings by 21σ to 341σ. The fifth — computational — was won as an instrument result (a sealed 40-bit answer decoded blind through noise, graded by three independent seats), then superseded by our own red-team before submission: the runtime number was priced against simulating the circuit, and a classical solver attacking the planted problem's algebra reads the same answer in ~0.25 ms.

The win's printed fence said exactly this could happen — supersedable by design: any classical method beating 1,818 s on the family retires the number. We fired that attack at our own instance first, and it did. The page honors its own retirement clause; the decoder physics (F120) stands untouched.

Read the Full Spec Sheet plain-text (.md)

The five scoreboards

Games & correlationsFOUND
216.8σ causal-order game
The three great no-go theorems in one court — Bell, indefinite causal order, contextuality (8/9 at 196σ). 216.8σ is within-run precision; the physical carrier is the two-chip replication (ibm_fez, ~34σ concordance). F82 · F106
Communication & sensingFOUND
341σ superdense coding
A ladder: superdense, 2→1 QRAC (110σ), GHZ metrology beats the standard quantum limit (168σ). F87 · F107 · F108/9
ThermodynamicsFOUND
21.1σ ICO refrigeration
The causal-indefiniteness engine: population inversion from passive baths, a full cycle, negative local energy. F86 · F94/5 · F97
InformationFOUND
42σ S(B|A) < 0
Negative conditional entropy — a sign classical physics forbids — plus zero-capacity transmission. F103 · F105
Computational⊘ SUPERSEDED
476× won, then retired by our own red-team
The sealed answer was read exactly in 3.82 s — the decoder physics is real (F120, instrument result). But the 1,818 s "classical floor" priced simulation; the problem's algebra falls to a 41-query linear-structure solve in ~0.25 ms (C4996, run on our own instance pre-submission). Runtime advantage retired; F119's sample-complexity claim is superseded as-executed too (own audit C4215 — delivery artifact, open floor; a conditional copies-edge is the residual). F119 · F120 · F121

Clearance over the classical / theorem bound · log scale

Through-line

The classical hardness of the computational bridge F113 is the magic-square contextuality F106, certified at 196σ — and the decoder race's shot-axis code F120 is temporal redundancy read from data the modal observable threw away. The correlation advantage and the computational instrument are the same resource, spent two ways.

What is not claimed

  • The "not a hardness theorem" fence fired — this is now the headline outcome. The 476× was priced against the best-known fielded classical simulation; the fence warned a solver attacking the planted problem's algebra directly could retire it. It did: the Maiorana–McFarland linear-structure attack (fixed-x linearity in y leaks the shift) recovers the sealed answer in 41 oracle queries / ~0.25 ms — run by our own red-team, verified independently by two more seats, before any submission. No runtime advantage is claimed. F121 · C4996
  • No moving bar — the WIN threshold (exact recovery + wall ≤ 1/10 of the band at every edge) was frozen word-for-word across a prior pre-registered miss (Race 5); only device-hygiene fences changed between attempts. The bar never moved — and the supersession came from the problem's algebra, not from relitigating the graded race.
  • One family, one die — Maiorana–McFarland n=40, t=80, ibm_kingston, one calibration window; joules one-sided.
  • No brute-force speedup — F54's 10× depth wall, measured; the decoder race is a different task and leaves it standing.
  • No on-chip QNC⁰≠NC⁰ proof — the separation is asymptotic; the apparatus is real.
  • No loophole-free device-independence — the randomness certificate (F117, 0.65 bit/use) is one-sided-DI; locality open.
  • Off-Heron, one axis only — the causal witness is the lone scoreboard tested beyond Heron, and it certifies there: cross-vendor (Rigetti Cepheus-1-108Q, W 1.2165, 54.4σ) and cross-modality (IonQ Forte-1 trapped ion, W 1.910 at 13.5σ, with a validated definite-order null). The other four scoreboards' numbers are same-Heron-generation measurements. F112 · Exp210/212