Skip to content

CONTRIBUTING.md

Latest commit

 

History

History
95 lines (66 loc) · 3.6 KB

CONTRIBUTING.md

File metadata and controls

95 lines (66 loc) · 3.6 KB

Contributing

All contributions are welcome. This dataset is being fact-checked file by file — every correction, addition, or pair of eyes helps.

No CLA, no contributor agreement. Public regulatory data, open contribution.

What's most useful

In priority order:

  1. Fact-corrections — you spotted a wrong allocation, power limit, modulation, or regulatory reference for your country
  2. Country additions — currently 9 covered (FR, US, UK, CN, DE, RU, ES, IT, CH). Adding JP, KR, BR, IN, AU, CA, others = huge value
  3. Device examples with model numbers — real-world devices that actually transmit on a band in your region
  4. Regulatory reference enrichment — pin a specific FCC Part / ECC Decision / national gazette to an entry
  5. HF datasets to cross-reference — point at community RF / rtl_433 datasets the pipeline should triangulate against
  6. Pipeline improvements — new sources, better confidence weighting, additional cross-check backends

Even a single-line correction is valuable — that's one less hallucination in someone's fine-tuned LLM.

How to contribute

Issues

Open one for:

  • Factual errors (please include the source: regulator URL, doc reference, etc.)
  • Missing devices / services / countries
  • Pipeline bugs
  • Suggestions

Pull Requests

Data corrections

  1. Edit the relevant JSON in enriched_data/ or merged_dataset/
  2. Add a correction_note field referencing the source you fact-checked against
  3. Update the relevant report in factcheck_reports/ if applicable
  4. PR description: what changed, why, source consulted

Example minimal correction:

{
  "freq_low_mhz": 868.0,
  "freq_high_mhz": 868.6,
  "country_code": "DE",
  "service": "SRD",
  "regulatory_ref": "BNetzA Vfg 60/2019, ETSI EN 300 220-2",
  "correction_note": "Power limit corrected from 25 mW ERP to 25 mW ERP with ≤1% duty cycle (was missing duty cycle constraint). Source: BNetzA Allgemeinzuteilung Vfg 60/2019 §4.2."
}

Pipeline changes

  • PRs against Data_Process/scripts/
  • Keep cross-check sources public / API-free where possible (DDG fallback, Wikipedia, HF public datasets)
  • Add tests / smoke checks if you change confidence math
  • Update Data_Process/README.md if you change behavior

New countries

  1. Add country to the relevant 5 enriched JSON files (one entry per allocation × country)
  2. Cite the national regulator at minimum
  3. Add the regulator to STANDARDS.md
  4. Add to the CROSS_CHECK_SOURCES.websearch.trusted_domains list in 00_config.py if they have an official .gov / national TLD

Code style

  • Python 3.10+
  • No formatter enforced; PEP8-ish
  • Keep network calls behind env flags (e.g. CROSS_CHECK=1) so the pipeline stays offline-first

What I'll merge fast

  • Typo fixes ✅
  • Single-line factual corrections with a source ✅
  • New device examples ✅
  • New regulatory citations ✅

What I'll discuss first

  • Removing entries (might need a disputed_note instead)
  • Changing the confidence scoring weights
  • Adding new cross-check sources (good — let's pick the right ones together)
  • Schema changes (need to migrate existing files)

Credit

Commit history is the credit log. If you want a different name in commits, configure your git locally before pushing — I won't rewrite history.

Questions?

Open an issue with the question label. Discussion is welcome before opening a big PR.

Thanks for helping make this dataset more accurate. RF allocations are genuinely complex and per-country edge cases are everywhere — community knowledge is the only way this gets to the ×210 production target with real quality.