Production-Grade QEC Decoding for Python

QECTOR Decoder v3 is a Rust-core Python library of 25+ production and research quantum error correction decoder configurations — MWPM Blossom, Belief-Matching, BP-OSD, Union-Find, GPU batch and more — behind one consistent API. v1.0.0 is the first stable release, with API stability tiers, a qector CLI and qector-doctor diagnostic.

pip install qector-decoder-v3==1.0.0

import numpy as np
from qector_decoder_v3 import BlossomDecoder
decoder = BlossomDecoder([[0, 1], [1, 2], [2, 3], [3, 4]], n_qubits=5)
correction = decoder.decode(np.array([0, 1, 0, 0], dtype=np.uint8))

v1.0.0 — first stable release

• Semantic-versioning frozen; the public API is governed by documented stability tiers (Stable / Workload-sensitive / Experimental).
• Sinter entry points (qector_blossom, qector_belief, qector_unionfind, and more) and a qiskit-qec plugin registered — sinter.collect() works with no custom_decoders=.
• New decoder families: AmbiguityClusterDecoder, TwoStageDecoder, ColourCodeDecoder (opt-in cluster_bposd). Relay-BP schedules, CS-OSD(lambda, w) and LLR damping in BP-OSD.
• qector decode / qector bench / qector serve CLI and qector-doctor (15-check environment diagnostic).
• Last published verified benchmark set (v0.7.0): peak 11.5M shots/s, 54/54 points with zero unfaithful corrections.

The decoders

UnionFind, FastUnionFind, Blossom MWPM, SparseBlossom, BeliefMatching, BP-OSD (qLDPC), batch decoders (CPU / CUDA / OpenCL), AutoDecoder (7-tier fallback), plus colour-code, two-stage and ambiguity-cluster decoders.

Evidence

• PyPI package: qector-decoder-v3 1.0.0
• Artifacts and reproduction harness: github.com/GuillaumeLessard/qector-decoder
• Verified v0.7.0 set (REPORT.md, summary.json, benchmarks.csv, VERIFIED_APPLE_TO_APPLE_REPORT.pdf) published at /benchmarks/v0.7.0/. Pre-v0.7.0 comparison tables are formally withdrawn.