MWPM surface-code decoder (PyMatching) — official baseline
Official KetQat baseline: minimum-weight perfect-matching (MWPM) decoding of rotated surface-code memory experiments, using Stim for circuit-level depolarizing-noise sampling and PyMatching for decoding. This is the reference decoder every KetQat leaderboard entry is compared against: 24 hash-verified baseline runs sweep code distances d=3/5/7 across physical error rates from 0.001 to 0.02, showing textbook below-threshold error suppression and above-threshold inversion. Reproduce any point with `ketqat run examples/qec/surface-code-memory.yaml` from the ketqat-sdk repository — every run page includes the exact manifest, environment, and reproducibility hash used. See the live comparison at ketqat.com/leaderboard.
QEC metadata
- Code families
- rotated-surface-code
- Decoder family
- minimum-weight perfect matching
- Distances
- 3, 5, 7
- Noise models
- circuit-level-depolarizing
- Syndrome format
- Stim detection events, unpacked bits (one bool per detector), from the shared detector sample
- Interface
- pymatching.Matching.from_detector_error_model; decode / decode_batch
Source
Repository: https://github.com/ketqat/ketqat-sdk
Language: Python
License: Apache-2.0
Cite this
@misc{ketqat_qecmwpmsurfacecodedecoderpymatchingofficialbaseline_2026,
title = {MWPM surface-code decoder (PyMatching) — official baseline},
author = {KetQat maintainers},
year = {2026},
howpublished = {\url{https://ketqat.com/qec/mwpm-surface-code-decoder-pymatching-official-baseline}},
note = {Source: \url{https://github.com/ketqat/ketqat-sdk}}
}Benchmark runs
Verified run
surface-code-mwpm-baseline-d7-p02
surface-code-memory-mwpm · v0.1.0
Verified run
surface-code-mwpm-baseline-d7-p015
surface-code-memory-mwpm · v0.1.0
Verified run
surface-code-mwpm-baseline-d7-p01
surface-code-memory-mwpm · v0.1.0
Verified run
surface-code-mwpm-baseline-d7-p008
surface-code-memory-mwpm · v0.1.0
Verified run
surface-code-mwpm-baseline-d7-p005
surface-code-memory-mwpm · v0.1.0
Verified run
surface-code-mwpm-baseline-d7-p003
surface-code-memory-mwpm · v0.1.0
Verified run
surface-code-mwpm-baseline-d7-p002
surface-code-memory-mwpm · v0.1.0
Verified run
surface-code-mwpm-baseline-d7-p001
surface-code-memory-mwpm · v0.1.0
Verified run
surface-code-mwpm-baseline-d5-p02
surface-code-memory-mwpm · v0.1.0
Verified run
surface-code-mwpm-baseline-d5-p015
surface-code-memory-mwpm · v0.1.0
Verified run
surface-code-mwpm-baseline-d5-p01
surface-code-memory-mwpm · v0.1.0
Verified run
surface-code-mwpm-baseline-d5-p008
surface-code-memory-mwpm · v0.1.0
Showing 12 of 49. View all
Datasets derived from this artifact's runs
- Surface-code MWPM baseline: logical error rates — 1 version, checksummed, lineage names this artifact's runs.
Related artifacts
No relations recorded. Relations are typed links -- for example “decodes” or “compatible with” -- asserted by a person and, where possible, backed by evidence.
Related artifacts
- Both have completed runs on the same benchmark suite (surface-code-memory-decoder-comparison). This is measured, not claimed.
- Uses the same QEC code family: rotated-surface-code.
- Both have been measured at code distance 3, 5. Overlapping scale is what makes a comparison possible; it does not by itself make one valid.
- Assumes the same noise model: circuit-level-depolarizing.
- Shares tags surface-code and mwpm.
- Both have completed runs on the same benchmark suite (surface-code-memory-decoder-comparison). This is measured, not claimed.
- Uses the same QEC code family: rotated-surface-code.
- Both have been measured at code distance 3, 5. Overlapping scale is what makes a comparison possible; it does not by itself make one valid.
- Assumes the same noise model: circuit-level-depolarizing.
- Shares tags surface-code and mwpm.
- Uses the same QEC code family: rotated-surface-code.
- Assumes the same noise model: circuit-level-depolarizing.
- Shares tags surface-code and mwpm.
These artifacts are related, which is a judgement about their metadata. That is not the same as being comparable: whether two results may be ranked together is decided by the compatibility rules against the actual runs, and shared tags or a shared code family do not establish it.
Verification evidence
No verification evidence recorded.