QECTOR Mega 4.5.0 Verification Artifact - Hardware, Decoder and Circuit Level Benchmark Report

Abstract

QECTOR Mega 4.5.0 Verification Artifact Hardware, Decoder and Circuit Level Benchmark Report Version1.1 (supersedes 1.0), 4 October 2026 AuthorsiD01t Productions (ORCID 0009-0000-3465-3753)Tested and verified by: Guillaume Lessard and Giro Charest Boujaklian Related WorkQECTOR Reference Manual v1.0.0, DOI: 10.5281/zenodo.21941046 Keywordsquantum error correction, qLDPC codes, bivariate bicycle codes, BPOSD, beam search decoding, Wilson interval, CUDA decoding, reproducibility AbstractThis report documents the complete outputs of one verification session of the QECTOR quantum error correction decoder suite, executed on a Google Colab host with two virtual CPUs and a CUDA-capable GPU. Three bodies of evidence are reported: Circuit-level memory benchmark of the bivariate bicycle code [[72,12,6]] comparing seven decoder variants over five physical error rates in both the X and Z bases (70 result rows, 38,125 decoded shots, exact Wilson 95 % intervals, 210 paired McNemar comparisons). 68-test certification suite reporting 68/68 PASS (covering decoders, IonQ target codes, code constructions, GPU batch paths, and the Genus 5 [[832,10]] hypergraph product code). Independent audit recomputing every confidence interval, conservative error rate, and per-round error rate from raw shot counts, confirming 70/70 rows consistent. The certification report is the unmodified output of the run; its status column reads 68 of 68 PASS. Every gated shot satisfies syndrome faithfulness (H c = s over GF(2)). Disclosed limitations (explicitly stated):• CUDA results are identical to the CPU reference and therefore are not independent proof of GPU execution.• The GPU path was slower than the CPU path on the repetition workload.• Four self-test checks were skipped rather than passed.• The certified distance of the Genus 5 code is the interval 7 ≤ d ≤ 15 rather than an exact value. Resource TypeDataset and technical report (verification artifact) Software under testqector_mega 4.5.0 with native extension (runtime reported); decoder version 4.5.0 Benchmark scriptqector_bb_benchmark.py 4.1.0, profile colab_t4_full Certification suiteQECTOR_PRO_MASTER, quick mode, 68 tests Zenodo DOI 10.5281/zenodo.23147058 Files (main deposit files) QECTOR_Zenodo_Publication_Report.pdf (18-page technical report) QECTOR_Zenodo_Artifact_v1.1.zip (complete deposit package) pro_suite/qector_pro_report.json (unmodified certification output) benchmark_suite/bb_circuit_results_fixed.csv (benchmark data with corrected unquoted code field) benchmark_suite/bb_circuit_plot.png benchmark_suite/bb_circuit_plot_per_round.png selftest.json SHA256SUMS.txt README.md All files are reproducible on Linux with qector_mega 4.5.0 (native=True), Python 3.12.3, and a CUDA-capable GPU. The deposit includes every raw output required for full reproduction and independent verification.