README.md

metadata

license: gpl-3.0
language:
  - en
  - fr
pretty_name: KiCad 9+ Schematic Corpus — Copyleft subset
task_categories:
  - text-generation
tags:
  - kicad
  - eda
  - schematic
  - s-expression
  - electronics
  - hardware
  - copyleft
size_categories:
  - n<1K

KiCad 9+ Schematic Corpus — Copyleft subset

209 KiCad 9+ schematic samples (.kicad_sch S-expression format, version >= 20240722).

Copyleft / strong reciprocal licenses only: GPL-3.0 (169), CERN-OHL-S-2.0 (36), EUPL-1.2 (4).

This is the copyleft split of the original electron-rare/kicad9plus-sch-corpus (now deprecated). The split was done after legal audit revealed CC-BY-SA-4.0 incompatibility with these inputs. CC-BY-SA-4.0 is one-way compatible to GPLv3 only, never the reverse — so the aggregated work has been re-licensed to GPL-3.0-or-later, the most restrictive of the inputs (CERN-OHL-S-2.0 and EUPL-1.2 are explicitly compatible with GPL-3.0+ via FSF / EUPL appendix interoperability).

For permissive samples (Apache-2.0, MIT, CC0-1.0, CERN-OHL-P-2.0): see electron-rare/kicad9plus-permissive.

EU AI Act compliance (Template AI Office, July 2025)

General information

• Name: kicad9plus-sch-corpus (copyleft subset)
• Modality: text (KiCad S-expression source)
• Languages: English (technical), French (some title-block descriptions)
• Intended use: training / fine-tuning of code-style language models on KiCad 9 / KiCad 10 schematic generation tasks. Models trained on this dataset must comply with GPL-3.0-or-later obligations (source disclosure of weights, derivative training data, and inference code if redistributed).

Data sources

Publicly available datasets: None.

Web scraping: Yes — public GitHub repositories.

• Discovery method: gh search code "(kicad_sch (version 202X)" extension:kicad_sch
• Filtering: SPDX license detection of GPL-3.0, CERN-OHL-S-2.0, EUPL-1.2
• Per-sample provenance: see metadata.source_url, metadata.commit_sha, metadata.repo

Licensed data: None (no commercial / proprietary licenses).

Data processing

• Sparse-clone with gh repo clone --depth 1
• Per-file .meta.json sidecar with source_url, commit_sha, license_spdx, kicad_version
• Deduplication via SHA-256 of file contents
• Truncation at 8 KB to fit training context (marked in metadata when applicable)
• Validation via kicad-cli sch erc --format json --severity-all (99.6% pass rate on tested subset)
• All processing scripts available at https://github.com/ailiance/ailiance-bench/tree/main/scripts

Data characteristics

• Size: 209 samples, ~2.1 MB JSONL (raw .kicad_sch totals ~38 MB before chat-format wrapping)
• License mix (input):
  • GPL-3.0: 169 samples (80.9%)
  • CERN-OHL-S-2.0: 36 (17.2%)
  • EUPL-1.2: 4 (1.9%)
• KiCad version mix: 20240819 / 20240910 / 20241004 / 20241209 (KiCad 9 dev), 20250114 (KiCad 9 stable), 20250227 / 20250318 / 20250610 / 20250829 / 20250901 / 20250922 / 20251012 / 20251028 (later 9.x), 20260101 / 20260306 (KiCad 10)
• Source repos: 9 distinct upstream projects (full list in LICENSE_INVENTORY.md); largest contributors are jaguilar/kicad (133) and flaviens/kicad (34) — both KiCad demo / fork repositories under GPL-3.0.

Licenses applied

This dataset (the aggregated work) is released under GPL-3.0-or-later.

Per-sample original licenses are preserved in metadata.license_spdx and listed in LICENSE_INVENTORY.md. Downstream users MUST preserve attribution per sample and comply with the strongest applicable copyleft term (GPL-3.0-or-later for the aggregate).

Compatibility notes:

• CERN-OHL-S-2.0 -> GPL-3.0+: explicitly compatible (CERN-OHL-S §7 allows redistribution under GPL when combining with GPL works).
• EUPL-1.2 -> GPL-3.0+: compatible via the EUPL §5 / Appendix list (GPL-3.0 is a listed compatible licence).
• GPL-3.0 -> GPL-3.0+: trivially compatible.

Copyright considerations

• All sources are public GitHub repositories under copyleft licenses.
• The .kicad_sch files are treated as software source under their original licenses.
• Opt-out mechanism: contact c.saillant@gmail.com to remove specific samples; we will respect Article 4(3) DSM Directive opt-outs (TDM reservations).
• Reservations of rights: we honor robots.txt, noai / noimageai HTML meta tags, and TDM Reservation Protocol (TDMRep) signals when discoverable on source repositories.

Sample format

Each line is a chat-format JSON object:

{
  "messages": [
    {"role": "user", "content": "Generate a KiCad 9 schematic (titled '...', by ..., N components, libraries: ...). Use the standard S-expression format starting with `(kicad_sch ...)`."},
    {"role": "assistant", "content": "(kicad_sch\n\t(version 20250114)\n\t..."}
  ],
  "metadata": {
    "repo": "owner/name",
    "rel_path": "path/to/file.kicad_sch",
    "source_url": "https://github.com/owner/name/blob/<sha>/path/to/file.kicad_sch",
    "commit_sha": "<git sha>",
    "license_spdx": "GPL-3.0",
    "kicad_version": "20250114",
    "file_sha256": "<content sha>",
    "file_size_bytes": 12345,
    "downloaded_at": "2026-05-11T...",
    "compliance_notes": "...",
    "ia_act_status": "requires_review"
  }
}

Pipeline reproducibility

See https://github.com/ailiance/ailiance-bench/tree/main/scripts:

• kicad9plus_pipeline.sh, build_kicad9plus_dataset.py
• Original audit: https://github.com/ailiance/ailiance-bench/blob/main/docs/audit_kicad9plus.md

Citation

@dataset{electron_rare_kicad9plus_copyleft_2026,
  author    = {electron-rare},
  title     = {KiCad 9+ Schematic Corpus — Copyleft subset},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/electron-rare/kicad9plus-copyleft}
}