Dataset Viewer
Auto-converted to Parquet Duplicate
distance
int64
3
7
basis
stringclasses
2 values
orientation
stringclasses
13 values
rounds
int64
1
250
decoder
stringclasses
21 values
model
stringclasses
4 values
shots
int64
10k
50k
n_errors
int64
0
26k
ler
float64
0
0.52
⌀
ler_per_cycle
float64
0
0.5
⌀
decode_seconds
float64
0
37k
⌀
note
stringclasses
20 values
3
X
q10_7
1
beliefmatching
-
50,000
209
0.00418
0.00418
0.233009
null
3
X
q10_7
10
beliefmatching
-
50,000
2,811
0.05622
0.005929
36.567719
null
3
X
q10_7
13
beliefmatching
-
50,000
3,692
0.07384
0.006108
61.353025
null
3
X
q10_7
30
beliefmatching
-
50,000
9,427
0.18854
0.007827
274.259002
null
3
X
q10_7
50
beliefmatching
-
50,000
11,660
0.2332
0.006242
516.17238
null
3
X
q10_7
70
beliefmatching
-
50,000
14,675
0.2935
0.006277
790.262073
null
3
X
q10_7
90
beliefmatching
-
50,000
17,100
0.342
0.006359
1,059.727906
null
3
X
q10_7
110
beliefmatching
-
50,000
19,121
0.38242
0.006536
1,317.697345
null
3
X
q10_7
130
beliefmatching
-
50,000
20,381
0.40762
0.006453
1,571.289211
null
3
X
q10_7
150
beliefmatching
-
50,000
21,399
0.42798
0.006417
1,817.780303
null
3
X
q10_7
170
beliefmatching
-
50,000
22,371
0.44742
0.006581
2,065.613964
null
3
X
q10_7
190
beliefmatching
-
50,000
22,669
0.45338
0.006205
2,304.842796
null
3
X
q10_7
210
beliefmatching
-
50,000
23,393
0.46786
0.006492
2,555.186118
null
3
X
q10_7
230
beliefmatching
-
50,000
23,806
0.47612
0.006569
2,811.521762
null
3
X
q10_7
250
beliefmatching
-
50,000
23,811
0.47622
0.006055
3,062.30165
null
3
X
q10_7
1
beliefmatching
-
10,000
42
0.0042
0.0042
0.048349
null
3
X
q10_7
10
beliefmatching
-
10,000
572
0.0572
0.006038
7.177917
null
3
X
q10_7
13
beliefmatching
-
10,000
758
0.0758
0.006283
12.333607
null
3
X
q10_7
30
beliefmatching
-
10,000
1,623
0.1623
0.006498
49.637775
null
3
X
q10_7
50
beliefmatching
-
10,000
2,317
0.2317
0.006186
103.814739
null
3
X
q10_7
70
beliefmatching
-
10,000
2,889
0.2889
0.006121
157.548035
null
3
X
q2_7
1
beliefmatching
-
50,000
290
0.0058
0.0058
0.242346
null
3
X
q2_7
10
beliefmatching
-
50,000
2,970
0.0594
0.006284
32.892415
null
3
X
q2_7
13
beliefmatching
-
50,000
3,998
0.07996
0.006658
54.248293
null
3
X
q2_7
30
beliefmatching
-
50,000
8,183
0.16366
0.006565
230.721629
null
3
X
q2_7
50
beliefmatching
-
50,000
12,189
0.24378
0.006641
504.506858
null
3
X
q2_7
70
beliefmatching
-
50,000
15,299
0.30598
0.006716
776.680555
null
3
X
q2_7
90
beliefmatching
-
50,000
19,796
0.39592
0.008644
1,068.911628
null
3
X
q2_7
110
beliefmatching
-
50,000
19,503
0.39006
0.006838
1,318.417553
null
3
X
q2_7
130
beliefmatching
-
50,000
21,512
0.43024
0.007518
1,570.085397
null
3
X
q2_7
150
beliefmatching
-
50,000
21,807
0.43614
0.006813
1,824.495864
null
3
X
q2_7
170
beliefmatching
-
50,000
22,654
0.45308
0.006911
2,065.277736
null
3
X
q2_7
190
beliefmatching
-
50,000
23,073
0.46146
0.006699
2,329.261695
null
3
X
q2_7
210
beliefmatching
-
50,000
23,594
0.47188
0.006806
2,568.25405
null
3
X
q2_7
230
beliefmatching
-
50,000
23,947
0.47894
0.006838
2,808.317556
null
3
X
q2_7
250
beliefmatching
-
50,000
23,977
0.47954
0.006352
3,053.487578
null
3
X
q2_7
1
beliefmatching
-
10,000
55
0.0055
0.0055
0.047707
null
3
X
q2_7
10
beliefmatching
-
10,000
600
0.06
0.006351
6.735189
null
3
X
q2_7
13
beliefmatching
-
10,000
808
0.0808
0.006733
10.848277
null
3
X
q2_7
30
beliefmatching
-
10,000
1,586
0.1586
0.006319
45.776521
null
3
X
q2_7
50
beliefmatching
-
10,000
2,380
0.238
0.006421
98.842008
null
3
X
q2_7
70
beliefmatching
-
10,000
3,094
0.3094
0.006842
156.765925
null
3
X
q4_5
1
beliefmatching
-
50,000
305
0.0061
0.0061
0.241683
null
3
X
q4_5
10
beliefmatching
-
50,000
3,599
0.07198
0.007712
40.089811
null
3
X
q4_5
13
beliefmatching
-
50,000
4,731
0.09462
0.008004
67.975616
null
3
X
q4_5
30
beliefmatching
-
50,000
9,267
0.18534
0.007659
256.537797
null
3
X
q4_5
50
beliefmatching
-
50,000
13,765
0.2753
0.007935
529.840213
null
3
X
q4_5
70
beliefmatching
-
50,000
16,742
0.33484
0.00785
807.420132
null
3
X
q4_5
90
beliefmatching
-
50,000
20,490
0.4098
0.009424
1,076.300485
null
3
X
q4_5
110
beliefmatching
-
50,000
20,553
0.41106
0.007787
1,323.456172
null
3
X
q4_5
130
beliefmatching
-
50,000
21,711
0.43422
0.007741
1,582.615899
null
3
X
q4_5
150
beliefmatching
-
50,000
22,640
0.4528
0.007806
1,819.486485
null
3
X
q4_5
170
beliefmatching
-
50,000
23,053
0.46106
0.007452
2,066.404044
null
3
X
q4_5
190
beliefmatching
-
50,000
23,812
0.47624
0.007953
2,316.053862
null
3
X
q4_5
210
beliefmatching
-
50,000
24,071
0.48142
0.007778
2,557.469671
null
3
X
q4_5
230
beliefmatching
-
50,000
24,104
0.48208
0.007184
2,797.624505
null
3
X
q4_5
250
beliefmatching
-
50,000
24,616
0.49232
0.008283
3,044.223644
null
3
X
q4_5
1
beliefmatching
-
10,000
62
0.0062
0.0062
0.048464
null
3
X
q4_5
10
beliefmatching
-
10,000
756
0.0756
0.00813
8.034247
null
3
X
q4_5
13
beliefmatching
-
10,000
927
0.0927
0.007825
13.589046
null
3
X
q4_5
30
beliefmatching
-
10,000
1,876
0.1876
0.007778
51.636493
null
3
X
q4_5
50
beliefmatching
-
10,000
2,737
0.2737
0.007865
107.290095
null
3
X
q4_5
70
beliefmatching
-
10,000
3,342
0.3342
0.007823
162.040909
null
3
X
q4_9
1
beliefmatching
-
50,000
245
0.0049
0.0049
0.237243
null
3
X
q4_9
10
beliefmatching
-
50,000
3,485
0.0697
0.00745
35.921973
null
3
X
q4_9
13
beliefmatching
-
50,000
4,515
0.0903
0.007602
59.915041
null
3
X
q4_9
30
beliefmatching
-
50,000
9,469
0.18938
0.007871
247.913934
null
3
X
q4_9
50
beliefmatching
-
50,000
14,100
0.282
0.008233
524.571066
null
3
X
q4_9
70
beliefmatching
-
50,000
16,854
0.33708
0.007946
793.854762
null
3
X
q4_9
90
beliefmatching
-
50,000
21,295
0.4259
0.010495
1,088.784254
null
3
X
q4_9
110
beliefmatching
-
50,000
21,259
0.42518
0.00856
1,321.603435
null
3
X
q4_9
130
beliefmatching
-
50,000
21,512
0.43024
0.007518
1,564.594896
null
3
X
q4_9
150
beliefmatching
-
50,000
22,373
0.44746
0.007454
1,824.724904
null
3
X
q4_9
170
beliefmatching
-
50,000
23,448
0.46896
0.008108
2,074.791034
null
3
X
q4_9
190
beliefmatching
-
50,000
23,507
0.47014
0.007361
2,314.74334
null
3
X
q4_9
210
beliefmatching
-
50,000
24,258
0.48516
0.008305
2,567.467281
null
3
X
q4_9
230
beliefmatching
-
50,000
24,190
0.4838
0.0074
2,821.992133
null
3
X
q4_9
250
beliefmatching
-
50,000
24,650
0.493
0.008465
3,072.402954
null
3
X
q4_9
1
beliefmatching
-
10,000
49
0.0049
0.0049
0.048903
null
3
X
q4_9
10
beliefmatching
-
10,000
704
0.0704
0.00753
7.09192
null
3
X
q4_9
13
beliefmatching
-
10,000
842
0.0842
0.007042
11.900012
null
3
X
q4_9
30
beliefmatching
-
10,000
1,934
0.1934
0.008085
49.283903
null
3
X
q4_9
50
beliefmatching
-
10,000
2,897
0.2897
0.008586
105.98059
null
3
X
q4_9
70
beliefmatching
-
10,000
3,332
0.3332
0.00778
157.756048
null
3
X
q6_11
1
beliefmatching
-
50,000
187
0.00374
0.00374
0.233503
null
3
X
q6_11
10
beliefmatching
-
50,000
3,351
0.06702
0.007144
39.133411
null
3
X
q6_11
13
beliefmatching
-
50,000
4,704
0.09408
0.007953
66.644245
null
3
X
q6_11
30
beliefmatching
-
50,000
9,273
0.18546
0.007666
259.448913
null
3
X
q6_11
50
beliefmatching
-
50,000
13,358
0.26716
0.007584
535.933484
null
3
X
q6_11
70
beliefmatching
-
50,000
17,937
0.35874
0.008948
827.807239
null
3
X
q6_11
90
beliefmatching
-
50,000
21,508
0.43016
0.010817
1,096.76585
null
3
X
q6_11
110
beliefmatching
-
50,000
20,378
0.40756
0.007614
1,328.781213
null
3
X
q6_11
130
beliefmatching
-
50,000
22,285
0.4457
0.008466
1,594.976714
null
3
X
q6_11
150
beliefmatching
-
50,000
22,607
0.45214
0.00776
1,826.093574
null
3
X
q6_11
170
beliefmatching
-
50,000
23,484
0.46968
0.008176
2,083.511269
null
3
X
q6_11
190
beliefmatching
-
50,000
24,130
0.4826
0.00876
2,326.803815
null
3
X
q6_11
210
beliefmatching
-
50,000
23,977
0.47954
0.007552
2,573.807472
null
3
X
q6_11
230
beliefmatching
-
50,000
24,297
0.48594
0.007704
2,814.926533
null
3
X
q6_11
250
beliefmatching
-
50,000
24,440
0.4888
0.00754
3,068.690035
null
3
X
q6_11
1
beliefmatching
-
10,000
48
0.0048
0.0048
0.04687
null
End of preview. Expand in Data Studio

Ising sim2real — Decoder Benchmark Results

Evaluation results for a panel of open surface-code decoders run on real Google Willow hardware data and on synthetic circuit-level noise of rising fidelity. The question these results answer: does the cheap synthetic benchmark predict the real-hardware result?

This repo holds the outputs (LERs, per-shot outcomes, fitted noise models, figures). The inputs — ingested Willow detection events, circuits, and shipped DEMs — live in ShayManor/willow-surface-code-detection-events.

The core idea

Decoder papers benchmark on synthetic circuit-level noise, then report a winner. Real hardware carries leakage, crosstalk, drift, soft readout, and rare high-energy events that no Pauli simulator produces. So: rows = data sources, columns = decoders, cell = that decoder's logical error rate (LER). Rows are ordered by how close the synthetic noise is to the device:

rung what it is
uniform uniform depolarizing noise
si1000 the standard SI1000 circuit-level noise model
fit a 25-parameter circuit-level model fitted to the device
syndrome a DEM estimated directly from syndromes (arXiv:2606.11496)
real real Willow data — the reference every synthetic rung is scored against, not a rung

Same circuit on every rung; only the noise model used to simulate detection events changes.

Decoder panel: mwpm (PyMatching), mwpm-rl (real only — uses the shipped RL-optimized DEM), tesseract, bposd, bplsd, beliefmatching, and ising (NVIDIA's Ising pre-decoder in front of PyMatching, model column = fast R=9 or accurate R=13).

Data: rotated surface code, distances 3/5/7, X and Z memory, 14 code patches, round counts r1–r250. Real source: Google Willow below-threshold dataset (10.5281/zenodo.13273331).

Quick start

from datasets import load_dataset

ladder = load_dataset("ShayManor/ising-sim2real-results", "ladder")
real = ladder["real"].to_pandas()

Or straight from the CSVs, which is usually what you want for analysis:

import pandas as pd

rungs = ["willow_real", "willow_synth_uniform", "willow_synth_si1000",
         "willow_synth_fit", "willow_synth_syndrome"]
df = pd.concat([
    pd.read_csv(f"hf://datasets/ShayManor/ising-sim2real-results/results/{r}/eval_all.csv")
      .assign(source=r.replace("willow_", ""))
    for r in rungs
])

# Decoder ranking per source, at matched distance/rounds:
cell = df.query("distance == 7 and rounds == 30")
cell.groupby(["source", "decoder"]).ler_per_cycle.mean().unstack()

Schema

Every eval_all.csv (and every per-config shard) uses one schema, so any two are directly comparable. One row = one (config x decoder).

column meaning
distance code distance d (3, 5, 7)
basis X or Z memory
orientation code patch label on the chip (e.g. q6_7)
rounds number of QEC cycles
decoder mwpm, mwpm-rl, tesseract, bposd, bplsd, beliefmatching, ising
model Ising model variant (fast / accurate), else -
shots shots decoded
n_errors raw logical-error count — makes LER exact and enables binomial bootstrap
ler logical error rate = n_errors / shots
ler_per_cycle LER normalized per QEC cycle — the comparable number across round counts
decode_seconds wall-clock decode time (latency analysis)
note e.g. skipped: rounds<2, or Ising's R=9,rot=XV

ler is nan where a config was skipped (see note). Compare ler_per_cycle, not ler, unless you have matched rounds.

Layout

results/
  willow_real/            REAL Willow row — the reference.  (+ outcomes/, 492 npz)
  willow_synth_uniform/   ladder rung: uniform depolarizing   (+ outcomes/, 446 npz)
  willow_synth_si1000/    ladder rung: SI1000                 (+ outcomes/, 448 npz)
  willow_synth_fit/       ladder rung: 25-param fitted        (+ outcomes/, 448 npz)
  willow_synth_syndrome/  ladder rung: syndrome-estimated DEM (+ outcomes/, 448 npz)
  ladder.png              the central deliverable figure

  willow_real_evalset/       REAL Willow, extended long-round tail (r110-r250)  (+ outcomes/, 178 npz)
  willow_synth_fit_evalset/  FIT rung, extended/held-out run (r1-r250)          (+ outcomes/, 474 npz)

  fitted_noise_models/    the 25-param fits, one JSON per patch (d{D}_q{loc}.json)

  sensitivity/            RQ3: per-parameter sweep, each of 23 params overestimated 2x
    baseline/             unperturbed fitted model, for comparison
    p_<param>/            one dir per perturbed parameter
    models/ models_under/ the perturbed noise models themselves
    sensitivity.png
  sensitivity_under/      same sweep, 0.5x UNDERestimate
  sensitivity_r10/        same sweep at r10
  sensitivity_r50/        same sweep at r50
  twobytwo/               2x2 decomposition: S-<sampled>__P-<prior> separates
                          "wrong noise sampled" from "wrong prior given to decoder"

  bootstrap/              correlated-perturbation eval, 40 refit draws (draw_00..draw_39)
    eval/draw_NN/         each draw's full panel eval
    models/draw_NN/       each draw's refitted noise models
    covariance.png

  t4_checkpoints/         pre-decoder finetune checkpoints (see "negative result" below)
  t4_ladder.png

outcomes/*.npz are bit-packed per-shot logical-error indicators, keyed d{D}|{basis}|{patch}|r{rounds}|{decoder}|n{shots} (Ising's fast/accurate variant is in the filename, not the key). Unpack with np.unpackbits(arr)[:n]; the sum equals that row's n_errors. All five rungs ship outcomes/, so a joint / rank bootstrap against real data is reproducible from this repo — not just a per-cell binomial one. Verified end-to-end: for every rung and every decoder, the unpacked per-shot sum equals the published eval_all.csv n_errors (1790/1790 keys on the real row, 0 mismatches; both *_evalset dirs also 0 mismatches against their own merged CSV).

The *_evalset dirs

willow_real_evalset/ and willow_synth_fit_evalset/ are separate eval campaigns, kept out of the main rungs on purpose — they use different shot counts (5,000 / 20,000, not 50,000) and mostly cover the long-round tail (r110–r250) that the main run doesn't decode. Same 12-column schema, own merged eval_all.csv, own outcomes/. Do not concatenate them into willow_real / willow_synth_fit: willow_real_evalset is entirely new (distance, decoder, round) cells, but willow_synth_fit_evalset overlaps the fit rung on some cells with different shot counts, so a blind merge would put two logical-error counts on one cell. Use them for long-round scaling curves; use the main rungs for the ladder ranking.

What the data shows

Headline numbers, so you know what you're looking at. Kendall tau between each rung's decoder ranking and real Willow's:

rung d3 d5 d7
uniform 0.619 (p=.07) 0.714 (p=.03) 0.733 (p=.06)
si1000 0.905 0.905 1.000
fit 0.905 0.905 0.810
syndrome 1.000 0.905 0.905
  1. Uniform depolarizing noise is a clear, often not-significant outlier. Any circuit-level structure (si1000/fit/syndrome) lifts agreement to tau >= 0.81. The fix is "have a real circuit," not necessarily "fit the noise precisely."
  2. The ladder is not monotonic — the more interesting result. At d7, generic si1000 (tau=1.000) beats both data-calibrated rungs despite being ~70x off in absolute LER. Fit/syndrome get the magnitude right and lose the rank check. Root cause: d7 has only 1 real patch, so fit/syndrome are weakly conditioned — estimation ill-conditioning, not a ladder-design flaw.
  3. The flip is always the same pair: mwpm <-> bposd/bplsd (near-tied mid-pack). tesseract and beliefmatching show zero rank drift anywhere.
  4. RQ3: rankings are robust. Overestimating any single noise parameter 2x causes at most one adjacent swap (tau >= 0.80); 10/23 params cause zero reordering. Only Y-containing / off-diagonal Pauli channels (p_idle_spam_Y, p_idle_cnot_Y, p_cnot_ZY) move anything; diagonal / dephasing channels including p_meas_Z are inert.
  5. RQ4: Ising loses to classical MWPM on real hardware by ~3x per-cycle at d7 (Z: 0.0121 vs 0.0038 at r70), while matching or beating MWPM on synthetic data. A genuine sim2real gap, not a bug — the signature is a round-independent error floor that synthetic data lacks. One caveat remains open: the Willow XZZX -> CSS mapping could produce a similar fixed offset.
  6. Negative result (t4_checkpoints/): a same-size pre-decoder finetuned via device-distillation does not beat stock NVIDIA Ising (0.0184 vs 0.0164). Structural — the observable isn't a function of the syndrome alone.

Caveats — read before drawing conclusions

  • The top-level results/*.csv files are STALE and should not be used. classical_d{3,5}_*.csv, ising_*_d{3,5}_*.csv are from Jun 28 2026, before a pipeline bug was fixed, and report Ising near chance (d3/Z ising_fast LER=0.371 vs the correct 0.128). They also predate the n_errors column (11 cols, not 12). They're kept only for provenance. Use willow_real/ and the willow_synth_*/ rungs, which were all re-run on the fixed pipeline.
  • Shot counts vary by decoder on the real row. Most decoders decode 50,000 shots; beliefmatching decodes 10,000. Both the CSV n_errors and the outcomes/ npz reflect this consistently (the npz key self-encodes n{shots}), but a joint per-shot bootstrap across decoders must align on the common shot count (10,000) or handle beliefmatching separately.
  • bplsd has no shard at uniform/d7 (a 446-vs-448 coverage gap), so that cell's tau is computed over 6 shared decoders rather than 7 and is not strictly comparable to the d3/d5 numbers beside it. This is why uniform's tau appears to improve with distance.
  • The sweep CSVs do not name their own perturbation. Every eval_all.csv under sensitivity*/, twobytwo/, and bootstrap/eval/ uses the same 12 columns with no column identifying which parameter was perturbed (or which draw it is) — the directory name is the only carrier. So don't blind-glob them together; tag as you load:
    import glob, pandas as pd
    sens = pd.concat([
        pd.read_csv(f).assign(param=f.split("/")[-2])
        for f in glob.glob("results/sensitivity/*/eval_all.csv")
    ])   # compare param="baseline" against each perturbed param
    
    This is also why only the ladder config is declared for the dataset viewer: there, the split name carries the rung, so nothing is lost.
  • Match your comparisons. Rungs are only comparable at matched (distance, basis, rounds) and matched-prior decoding (each rung decodes with the DEM it sampled from). The headline table uses rounds 2–30.
  • mwpm-rl exists only on real (no synthetic rung has an RL-optimized DEM), so exclude it from cross-rung comparisons.
  • Always report code distance d alongside the Ising receptive field R (in note). The public Ising models were trained for larger R than d=3/5/7.

Reproducing

Harness: github.com/ShayManor/Ising-sim2real. Jobs produce raw data only; all statistics (Kendall tau, Spearman, bootstrap CIs) are computed locally from these CSVs via scripts/build_ladder.py, scripts/build_sensitivity.py, scripts/build_bootstrap.py, scripts/build_2x2.py, and scripts/build_selection.py.

Downloads last month
642

Paper for ShayManor/ising-sim2real-results