# QECTOR Decoder v3 (v1.0.0)

> Python and Rust quantum error correction decoder package for QEC research, decoder validation, reproducible artifacts, and commercial evaluation.

QECTOR Decoder v3 is documented by the QECTOR Decoder v3 Reference Manual v1.0.0, deposited at DOI https://doi.org/10.5281/zenodo.21941046. The manual is the source of truth for decoder contracts, evidence anchors, limitations, and claim boundaries.

The common correctness contract is syndrome faithfulness: for a reachable syndrome `s`, a returned correction `c` satisfies `H c = s (mod 2)`. Logical scoring is separate and must use observables or stabilizer cosets rather than raw correction equality.

Decoder families include weighted Blossom MWPM, event-driven Sparse Blossom, graphlike Union-Find, BP-OSD for arbitrary GF(2) check matrices, ambiguity clustering, two-stage CSS decoding, space-time decoding, streaming primitives, lookup tables, CPU/GPU batch paths, and native routing/orchestration. The actual check structure determines eligibility: mechanisms touching more than two detectors are hyperedges and route away from matching.

No universal latency, throughput, memory, VRAM, or threshold figures are published as QECTOR facts. The manual requires workload-specific metadata, raw JSON/CSV artifacts, SHA-256 hashes, and Wilson intervals for logical-error rates. Earlier hardware-bound benchmark artifacts that did not survive a core fingerprint change are withdrawn.

The public blog provides 20 field notes across foundations, decoder algorithms, qLDPC and AI, multi-round decoding, noise models, evidence, systems, orchestration, deployment, and ecosystem integration:

- https://qector.store/blog

## QECTOR Workbench (desktop GUI + MCP server)

- Windows v1.0.1: 85-tool MCP server, qector-decoder-v3 1.0.0 backend, 17 decoder kinds, 10 code families. Portable exe + SHA-256 checksums.
- Linux v1.0.1: 85-tool MCP server, 1.0.0 backend, 17 decoder kinds, 10 code families. AppImage + Debian packages.
- MCP protocol 2024-11-05 over stdio JSON-RPC 2.0 (`--mcp` headless mode); data directory `~/.local/share/QectorWorkbench` (`QECTOR_DATA_DIR` overrides).
- No macOS build is currently published.

## QECTOR Claude Plugin (v1.0.2)

- Claude Code / Claude Desktop plugin: 28 domain skills, 5 specialized agents (qec-developer, qec-researcher, qec-validator, qec-sysadmin, qec-hardware-engineer), 13 slash commands.
- 37 local MCP tools: 8 library (list_code_families, list_decoders, get_license_info, decode_syndrome, decode_single, threshold_sweep, build_code_from_matrix, compat_report) + 29 benchmark.
- Zero-egress stdio architecture: circuits, parity matrices, and syndromes never leave the machine.
- Install: `claude plugin marketplace add GuillaumeLessard/qector-claude-plugin` then `claude plugin install qector@qector-tools`.

## Official links

- Website: https://qector.store/
- Claude Code Plugin: https://qector.store/claude-plugin
- Workbench: https://qector.store/workbench
- MCP Server: https://qector.store/mcp-server
- Blog: https://qector.store/blog
- Master AI Suite v2: https://qector.store/master-ai-suite
- Pricing & licensing: https://qector.store/pricing
- Enterprise / OEM licensing: https://qector.store/commercial
- Install guide: https://qector.store/installer
- Documentation hub: https://qector.store/docs
- User manual: https://qector.store/manual
- Technical reference: https://qector.store/technical-reference
- Evidence and reports: https://qector.store/evidence
- About: https://qector.store/about
- Founder page: https://qector.store/guillaume-lessard
- Changelog: https://qector.store/changelog
- License: https://qector.store/license
- Refund policy: https://qector.store/refund
- Contact: https://qector.store/contact

## Machine-readable

- Full-context document: https://qector.store/llms-full.txt
- AI agents index: https://qector.store/ai.txt
- Atom feed (blog): https://qector.store/feed.xml
- Sitemap: https://qector.store/sitemap.xml
- Agent skills: https://qector.store/.well-known/skills/index.json
- MCP server card: https://qector.store/.well-known/mcp/server-card.json
- Security contact: https://qector.store/.well-known/security.txt

## External records

- DOI record: https://doi.org/10.5281/zenodo.21941046
- PyPI package: https://pypi.org/project/qector-decoder-v3/
- Primary GitHub repository: https://github.com/GuillaumeLessard/qector-decoder
- Workbench Windows releases: https://github.com/qectorlab/qector-decoder-workbench-windows/releases
- Workbench Linux releases: https://github.com/qectorlab/qector-decoder-workbench-linux/releases
- Claude plugin repository: https://github.com/GuillaumeLessard/qector-claude-plugin
- Author ORCID: https://orcid.org/0009-0000-3465-3753

## Installation

```text
pip install qector-decoder-v3==1.0.0
```

## Reproducibility

The manual provides build/import smoke commands, focused correctness tests, LER workflows, GPU identity workflows, and SHA-256 artifact hashing. A benchmark must record code family, size, noise model, decoder configuration, shots, seed, environment, and raw artifact path.

## Boundaries and licensing

- Free academic, personal, and non-commercial research use is governed by the published project license.
- Commercial production, hosted API, OEM, redistribution, and commercial benchmarking require the written commercial scope described by QECTOR.
- Local CPU decoding is the preferred research path.
- CUDA/OpenCL, REST, gRPC, MCP, and metrics paths require the hardware, feature, and deployment review described by the manual.