Independent evaluation · Quantum computing
Quantum, independently measured.
Pre-registered, sealed, published win or lose — so you can build or invest on evidence.
Survival clock
what happens to quantum claims after the press release · bar = share of each claim's life survived unchallenged
Quantum supremacy — Google Sycamore, 2019 · eroded
Quantum utility — IBM Eagle, 2023 · eroded
Quantum Echoes (OTOC) — Google, 2025 · contested
Below-threshold error correction — Google Willow, 2024 · surviving
Our own quantum walk — RQC-2026-ROSETTA-QMARGIN ours · negative, self-published
■ surviving · ■ contested · ■ eroded or self-negative
all 16 tracked claims
The Evidence Library.
Is there real evidence for this quantum claim?
A search engine for quantum computing evidence — every claim dated, sourced, statused, and linked to its seal. Of 16 tracked advantage claims: 3 surviving, 5 contested, 6 eroded, 1 open, 1 self-published negative.
certified randomness
Willow error correction
QAOA portfolio
Sycamore supremacy
fraud detection kernel
74
algorithms catalogued
625
references indexed
16
advantage claims tracked
3
still surviving · 19%
Free to browse · Analyst US$149 per seat / month · everything else on request — evidence levels identical in every tier
The fight, shown. Not described.
Five industries, five sealed experiments, one rule: the classical champion runs on the same field. Pick one.
KRAS G12C · PDB 4OBE · 169 residues · 927 contacts (Cα–Cα ≤ 8.5 Å) · sealed challenge cleveland-2026-07
iψ̇ = Hψ integrated by RK4 in your browser · t = 0.00 · not a video
◉ 25 source residues · ○ 20 true allosteric residues (ground truth, hidden from the method)
In plain termsWe tested whether a "quantum walk" finds the spots where a drug can grab a cancer protein better than classical methods. It didn't just lose to all five classical methods — it lost to picking at random. We published it anyway, because it's our own experiment.
Read this as a table
| Method | Mean percentile of the true allosteric residues |
|---|---|
| Continuous-time quantum walk (CTQW) | 35.63 |
| Classical diffusion | 53.65 |
| Random | 49.38 |
Higher is better: the true residues should rank near the top. Sealed as EXP-0007-013.
EXP-0007-013 · outcome: quantum worse than random · mean percentile CTQW 35.63 · diffusion 53.65 · random 49.38 · verdict: pending
The live walk runs the same code as our 3D piece (H = gaussian-weighted Cα contacts, cutoff 8.5 Å, σ 6.0, dt 0.01). Checked against the sealed result: Spearman ρ = 0.746 on the ranking; normalization differs (declared, not adjusted). The sealed run is the record; the live walk is the method, shown.
sealed runchallenge data
V-0012 · Constrained portfolio compression · pick k of n assets · QAOA p=2 vs OR-Tools CP-SAT · 20 sealed runs
every asset, edge and choice below is regenerated from the sealed instance — not an illustration
EXP-0012-001 · ■ optimum · ○ QAOA · edges = |correlation| ≥ 0.25
In plain termsWe put a quantum algorithm and a classical one on the same investment problem, same time, same machine. The classical one found the perfect answer in all twenty tests; the quantum one always fell short — between 10% and 87% depending on the seed — and the gap does not shrink consistently as the problem grows. The 20 instances were regenerated from the sealed seeds and re-solved by brute force: all 20 optima match the seal exactly.
RQ-0033 · Grid expansion · IEEE case118 · 118 buses · 173 lines · choose 5 of 14 candidate lines · seed 42the real network from the sealed topology (positions are a layout, not geography) — not an illustration
┈ 14 candidate lines · ━ the 5 built
In plain termsChoose which 5 new lines to build in a 118-bus power grid to carry more load at the lowest cost. The classical solver finds the exact optimum in 0.2 seconds; the simulated quantum one lands 1% above it after 52 seconds; and the run on a real IBM quantum chip didn't return enough valid answers to compare at all.
eon_case118@a3340c06 · outcome: not yet — classical wins · QPU run RQ-EON-QPU-001 · ibm_marrakesh · 2,196 of 8,192 answers valid (26.8%) vs 24.6% blind chance · verify ↗sha256 6781f2d0…e12c simulated runQPU runraw counts
RQ-EXP-HSBC-Q-001 · Fraud detection · 56,962 real card transactions, 75 frauds · quantum fidelity kernel vs classicalevery dot is a sealed score — not an illustration
● the 75 frauds · · 3,000 of 56,887 legitimate
In plain termsA quantum kernel tried to spot fraudulent transactions better than an already-trained classical model. It lost: 25.7% precision-recall against 80.1%. We checked whether having fewer examples to learn from was the excuse — it wasn't: a simple classical model with the same handicap still tied the full classical baseline.
RQ-EXP-HSBC-Q-001 · sealed 2026-08-26 · verify ↗sha256 436334b4…075f · quantum AUPRC 0.257 [0.155–0.369] · classical baseline (XGBoost, sealed) 0.801 [0.705–0.883] · outcome: no advantage (intervals do not overlap) sealed runscores
Both score files are the sealed ones: quantum kernel sha256 091914f1… · classical XGBoost baseline sha256 62c29285… (RQ-EXP-HSBC-BASE-001). AUPRC recomputed from the files: 0.257453 and 0.800822 — identical to the seals. Temporal 80/20 split, no exact duplicates across halves.
Airbus challenge · 2-D incompressible flow · Taylor–Green vortex · classical sweep Re 10 → 102,400 vs quantum Carleman armdrawn from the sealed raw data (hash-verified) — not an illustration
In plain termsThe challenge asks solvers to handle the hard, nonlinear part of fluid flow. We measured two things. Across the full sweep, the classical solver’s error falls four orders of magnitude while the quantum approach never enters the field: at every point it needs more qubits than its own declared budget — 21 to 45 against a cap of 12. And in the exact vortex the statement specifies, the nonlinear term cancels to machine zero — so the benchmark cannot tell a correct solver from one that skips the physics entirely.
RQ-PREREG-AIRBUS-001 · 4 sealed runs, 4 raw artifacts by hash · outcome: quantum arm did not qualify · declared advantage crossings: 0 · verify ↗sha256 392c8f1a…4ab0 · 7a3aaf38…ab67 sweep runnonlinearity runraw data
▸ every panel is drawn live from the sealed data — drag, switch and replay · the full fight & the industry cases
RQ Advantage Monitor · Edition 001 · in preparation
What the ledger can already say.
3 / 16Still standing. Of 16 tracked advantage claims since 2009, three survive unchallenged; six have eroded under later work.
2 daysFastest challenge. Google's 2019 supremacy claim drew its first serious challenge within 2 days; IBM's 2023 utility claim, within 5.
oursThe only claim open with zero challenges is our own — N=90 allosteric significance. We are waiting to be checked like everyone else.
EL3EL3 — independently reproduced — is empty. Across the whole field, no third party has ever confirmed another's headline result from its published artifacts. That gap is the product.
In plain terms: most public quantum-advantage claims are challenged within weeks; the few that stand for years are about error correction and hardware — the foundations the useful applications will be built on.
one email when the edition seals · no list, no drip
If a vendor shows you the animation, ask for the last two numbers.
With a referee, and without one.
Without an independent referee
- The vendor picks the baseline and the compute budget.
- The success criterion is written after the results are in.
- A negative result never gets published.
- Your committee gets a PDF it cannot audit.
With one
- The criterion is registered and sealed before the run.
- The classical champion runs at the same budget.
- The result is published either way, hash-anchored, with a date.
- Your committee gets a decision page with an ID it can cite.
Latest from the desk.
Maps & data · 2026-08-31
Neutral atoms: is the third architecture for real?
As science, measurably yes: the largest fault-tolerant demonstration published to date (448 atoms) and the best physical-to-logical ratio on any platform. As product, the numbers say something else.
Pillar E · Maps & data · 6 min read
Canonical dictionary · 2026-08-30
What is QRAM, and why is it the silent bottleneck?
The device that would let a quantum computer read classical data in superposition — and the unstated hardware assumption behind most quantum machine-learning claims. It does not exist yet.
Pillar B · Canonical dictionary · 5 min read
93 sealed runs · 0 quantum wins measured.
Every experiment, report and correction we produce is hash-anchored to Bitcoin via OpenTimestamps and mirrored on GitHub, Codeberg and D1 — including the ones we lose.
Open the full ledger