The evidence ledger

Every verdict, with the raw proof.

The public record of Rosetta's verdicts: per recipe and problem class, whether quantum beat the best classical solver, at what size, with reproducible evidence.

4
Recipes in pipeline
48
Sealed and anchored runs
1
Verdicts published
3×
Copies per run + external anchor
ID
Recipe
Problem class
Vertical
Status
Evidence
RQ-0007
Allosteric site detection by signal propagation
Allosteric site prediction
Pharma
measuring…
view detail →
RQ-0012
Constrained portfolio compression
Portfolio optimization
Finance
measuring…
view detail →
RQ-0019
Fleet routing under uncertainty
Vehicle routing
Mining
in test
soon
RQ-0033
Grid expansion under stress
Grid optimization
Energy
measuring…
view detail →
RQ-0007 · Allosteric site detection by signal propagation
Algorithm
Continuous-time quantum walk (CTQW) vs classical diffusion
Problem class
Allosteric site prediction
Qubits (est.)
9
Source
arXiv (harvested)
No verdict: the data is measured and sealed, but we are not calling a result yet.
Instance
Quantum side
Classical champion
Seed
Evidence
2HHB c8.5 w0.5-8.0 (n=574)
CTQW exp(-iAt), exact matrix-exp · allosteric pct 69.89
diffusion exp(-Lt), exact matrix-exp · allosteric pct 74.24
deterministic
2HHB c7.5 w0.5-8.0 (n=574)
CTQW exp(-iAt), exact matrix-exp · allosteric pct 57.32
diffusion exp(-Lt), exact matrix-exp · allosteric pct 73.87
deterministic
2HHB c8.0 w0.5-8.0 (n=574)
CTQW exp(-iAt), exact matrix-exp · allosteric pct 75.67
diffusion exp(-Lt), exact matrix-exp · allosteric pct 73.65
deterministic
2HHB c9.0 w0.5-8.0 (n=574)
CTQW exp(-iAt), exact matrix-exp · allosteric pct 79.08
diffusion exp(-Lt), exact matrix-exp · allosteric pct 76.82
deterministic
2HHB c9.5 w0.5-8.0 (n=574)
CTQW exp(-iAt), exact matrix-exp · allosteric pct 56.83
diffusion exp(-Lt), exact matrix-exp · allosteric pct 77.31
deterministic
2HHB c10.0 w0.5-8.0 (n=574)
CTQW exp(-iAt), exact matrix-exp · allosteric pct 71.82
diffusion exp(-Lt), exact matrix-exp · allosteric pct 76.57
deterministic
2HHB c8.5 w0.5-4.0 (n=574)
CTQW exp(-iAt), exact matrix-exp · allosteric pct 73.4
diffusion exp(-Lt), exact matrix-exp · allosteric pct 71.97
deterministic
2HHB c8.5 w0.5-14.0 (n=574)
CTQW exp(-iAt), exact matrix-exp · allosteric pct 74.21
diffusion exp(-Lt), exact matrix-exp · allosteric pct 76.23
deterministic
1SHJ c8.0 w0.5-8.0 (n=417)
CTQW exp(-iAt), exact matrix-exp · allosteric pct 70.87
diffusion exp(-Lt), exact matrix-exp · allosteric pct 73.98
deterministic
1SHJ c8.5 w0.5-8.0 (n=417)
CTQW exp(-iAt), exact matrix-exp · allosteric pct 60.26
diffusion exp(-Lt), exact matrix-exp · allosteric pct 70.19
deterministic
1SHL c8.5 w0.5-8.0 (n=398)
CTQW exp(-iAt), exact matrix-exp · allosteric pct 72.7
diffusion exp(-Lt), exact matrix-exp · allosteric pct 72.87
deterministic
1B86 c8.5 w0.5-8.0 (n=574)
CTQW exp(-iAt), exact matrix-exp · allosteric pct 33.69
diffusion exp(-Lt), exact matrix-exp · allosteric pct 25.36
deterministic
KRAS G12C · 4OBE -> 6OIM chain A · 18-config grid
CTQW exp(-iAt), 18 configs · mean pct 35.63 (top-5 hits 5/18)
6 classical baselines (diffusion/GNM/ANM/betweenness/closeness/random) · best mean pct 61.68 (closeness); random 49.38
deterministic
BCR-ABL1 · 1OPL -> 5MO4 chain A · 18-config grid
CTQW exp(-iAt), 18 configs · mean pct 57.99 (top-5 hits 0/18)
6 classical baselines (diffusion/GNM/ANM/betweenness/closeness/random) · best mean pct 61.9 (betweenness); random 48.28
deterministic
cardiac myosin · 5TBY -> 9GZ1 chain A · 18-config grid
CTQW exp(-iAt), 18 configs · mean pct 63.86 (top-5 hits 0/18)
6 classical baselines (diffusion/GNM/ANM/betweenness/closeness/random) · best mean pct 58.02 (diffusion); random 47.06
deterministic
c-Myc/Max · 1NKP chains A+B (c-Myc/Max on E-box DNA) · blind prediction
CTQW, top-5 sites per config · sealed and dated 2026-07-25; no co-crystallised effector exists to score against
diffusion exp(-Lt) · same blind protocol; not scorable today
deterministic
contiguous-pocket spatial null · KRAS G12C, BCR-ABL1, cardiac myosin
our own method audited first: CTQW · not significant on any target (p in [0.15, 0.85])
same null applied to all 6 classical baselines · not significant either; naive i.i.d. z up to |4.64| collapses to |z| < 1.2 (2000 perms)
20260717
pocket crypticity · KRAS G12C (4OBE/6OIM), BCR-ABL1 (1OPL/5MO4)
not applicable: structural diagnosis, not a propagator · measures why the prediction fails, it does not compete
Ca apo-holo displacement + contact counts · two failure regimes: cryptic (KRAS, site/rest ratio 2.05) vs preformed-rigid (BCR-ABL1, 0.59)
deterministic
noise, scale and circuit cost · KRAS G12C, BCR-ABL1, cardiac myosin
Trotterised CTQW, binary encoding · 8-9 qubits; r=256 Trotter steps for Spearman 0.997 vs exact; 2q-depth 4864 (KRAS)
exact eigendecomposition propagator as reference · dephasing gamma=1.0 still Spearman 0.85-0.91; coarse-graining 12x-42x faster with percentile drift
deterministic
RQ-0012 · Constrained portfolio compression
Algorithm
QAOA
Problem class
Portfolio optimization
Qubits (est.)
20
Source
arXiv (harvested)
not yet
CP-SAT reached the proven exact optimum in all 20 sealed runs (n=12: 8 seeds · n=16: 8 · n=20: 4). QAOA p=2 mean gaps to optimum: 48.2% / 25.1% / 41.1% under equal 120 s budgets. Sealed verdict V-0012, sha256:f510eff6…6636, triple copy + OTS.
Crossover not observed — no defensible size trend at n≤20; seed-to-seed variance (σ 11–23 pts) is first-order. Series continues.
Instance
Quantum side
Classical champion
Seed
Evidence
portfolio_12_assets_seed42
PennyLane QAOA p=2 (CPU sim) · -0.3350 (gap 42.8%)
OR-Tools CP-SAT · -0.5858 (proven optimal)
42
portfolio_16_assets_seed42
PennyLane QAOA p=2 (CPU sim) · -0.4325 (gap 20.7%)
OR-Tools CP-SAT · -0.5457 (proven optimal)
42
portfolio_12_assets_seed43
PennyLane QAOA p=2 (CPU sim) · -0.1304 (gap 53.7%)
OR-Tools CP-SAT · -0.2818 (proven optimal)
43
portfolio_12_assets_seed44
PennyLane QAOA p=2 (CPU sim) · -0.2310 (gap 43.7%)
OR-Tools CP-SAT · -0.4104 (proven optimal)
44
portfolio_12_assets_seed45
PennyLane QAOA p=2 (CPU sim) · -0.3704 (gap 23.4%)
OR-Tools CP-SAT · -0.4835 (proven optimal)
45
portfolio_12_assets_seed46
PennyLane QAOA p=2 (CPU sim) · -0.3166 (gap 39.8%)
OR-Tools CP-SAT · -0.5264 (proven optimal)
46
portfolio_12_assets_seed47
PennyLane QAOA p=2 (CPU sim) · -0.3375 (gap 22.0%)
OR-Tools CP-SAT · -0.4328 (proven optimal)
47
portfolio_12_assets_seed48
PennyLane QAOA p=2 (CPU sim) · -0.0855 (gap 73.4%)
OR-Tools CP-SAT · -0.3220 (proven optimal)
48
portfolio_12_assets_seed49
PennyLane QAOA p=2 (CPU sim) · -0.0417 (gap 86.5%)
OR-Tools CP-SAT · -0.3097 (proven optimal)
49
portfolio_16_assets_seed43
PennyLane QAOA p=2 (CPU sim) · -0.3160 (gap 42.3%)
OR-Tools CP-SAT · -0.5480 (proven optimal)
43
portfolio_16_assets_seed44
PennyLane QAOA p=2 (CPU sim) · -0.3494 (gap 21.3%)
OR-Tools CP-SAT · -0.4438 (proven optimal)
44
portfolio_16_assets_seed45
PennyLane QAOA p=2 (CPU sim) · -0.3207 (gap 31.0%)
OR-Tools CP-SAT · -0.4649 (proven optimal)
45
portfolio_16_assets_seed46
PennyLane QAOA p=2 (CPU sim) · -0.4116 (gap 31.2%)
OR-Tools CP-SAT · -0.5982 (proven optimal)
46
portfolio_16_assets_seed47
PennyLane QAOA p=2 (CPU sim) · -0.2926 (gap 32.7%)
OR-Tools CP-SAT · -0.4349 (proven optimal)
47
portfolio_16_assets_seed48
PennyLane QAOA p=2 (CPU sim) · -0.3965 (gap 12.3%)
OR-Tools CP-SAT · -0.4520 (proven optimal)
48
portfolio_16_assets_seed49
PennyLane QAOA p=2 (CPU sim) · -0.5060 (gap 9.6%)
OR-Tools CP-SAT · -0.5598 (proven optimal)
49
portfolio_20_assets_seed42
PennyLane QAOA p=2 COBYLA (CPU sim) · -0.5794 (gap 24.3%)
OR-Tools CP-SAT · -0.7648 (proven optimal)
42
portfolio_20_assets_seed43
PennyLane QAOA p=2 COBYLA (CPU sim) · -0.2021 (gap 58.5%)
OR-Tools CP-SAT · -0.4869 (proven optimal)
43
portfolio_20_assets_seed44
PennyLane QAOA p=2 COBYLA (CPU sim) · -0.3462 (gap 42.9%)
OR-Tools CP-SAT · -0.6065 (proven optimal)
44
portfolio_20_assets_seed45
PennyLane QAOA p=2 COBYLA (CPU sim) · -0.3994 (gap 38.8%)
OR-Tools CP-SAT · -0.6528 (proven optimal)
45
RQ-0033 · Grid expansion under stress
Algorithm
QAOA p=2 vs CP-SAT
Problem class
Grid optimization
Qubits (est.)
16
Source
arXiv (harvested)
No verdict: the data is measured and sealed, but we are not calling a result yet.
Instance
Quantum side
Classical champion
Seed
Evidence
case14_stress3.0_K14_seed42
PennyLane QAOA p=2 (CPU sim, 2000 shots) · 3493.43 (gap 3.20%)
OR-Tools CP-SAT · 3384.95 (proven optimal)
42
case14_stress3.0_K14_seed43
PennyLane QAOA p=2 (CPU sim, 2000 shots) · 3385.12 (gap 0.01%)
OR-Tools CP-SAT · 3384.95 (proven optimal)
43
case14_stress3.0_K14_seed44
PennyLane QAOA p=2 (CPU sim, 2000 shots) · 3420.02 (gap 1.04%)
OR-Tools CP-SAT · 3384.95 (proven optimal)
44
case14_stress2.2_K16_seed42
PennyLane QAOA p=2 (CPU sim, 2000 shots) · 11638.15 (gap 1.04%)
OR-Tools CP-SAT · 11518.69 (proven optimal)
42
case14_stress2.2_K16_seed43
PennyLane QAOA p=2 (CPU sim, 2000 shots) · 9274.08 (gap 1.03%)
OR-Tools CP-SAT · 9179.27 (proven optimal)
43
case14_stress2.2_K16_seed44
PennyLane QAOA p=2 (CPU sim, 2000 shots) · 8034.33 (gap 0.99%)
OR-Tools CP-SAT · 7955.70 (proven optimal)
44
case14_stress2.2_K16_seed45
PennyLane QAOA p=2 (CPU sim, 2000 shots) · 3583.37 (gap 3.11%)
OR-Tools CP-SAT · 3475.40 (proven optimal)
45
case14_stress2.2_K16_seed46
PennyLane QAOA p=2 (CPU sim, 2000 shots) · 9109.95 (gap 1.45%)
OR-Tools CP-SAT · 8979.99 (proven optimal)
46
case14_stress1.8_K16_seed42
PennyLane QAOA p=2 (CPU sim, 2000 shots) · 2863.35 (gap 4.20%)
OR-Tools CP-SAT · 2747.96 (proven optimal)
42

We audit ourselves

A published seal is never rewritten. When we find our own error, we publish the correction as a new file that cites the original by its hash — the original stays intact and checkable.

reading /v1/erratas…