- data/dictionary.yaml: every published file, and every column of runs_clean.csv with its type, meaning and allowed values - collection/publish_metadata.py writes data/croissant.json (Croissant 1.0) from it, with a sha256 per file; `npm run metadata`, run by the daily update - tests/core/test_dictionary.py: every column and value described, types hold, croissant.json up to date and checksums match - attempt is now written as a whole number (was 1.0, 2.0); no statistic changes - the site ships both files and the Methods text links them Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01UCzTd31GphpMXDhQqWrCMQ
collection
Getting data in. Everything here turns a source into the shared run format (see core/) or into public files
under data/.
| Source | Code | Output |
|---|---|---|
| Community "Threshing Routes" form (Google Sheet) | sources/export_raw.py, then core/atlas/coding.py |
data/public/form_responses.json, data/derived/runs_clean.csv |
| Unofficial Dragon Register (Google Sheet) | sources/export_register.py |
data/public/register.json (rider names stay local) |
| Community route PDF | sources/update_sources.py |
data/derived/pdf_fingerprint.json only; the PDF is never committed |
| DragonKind bot (Ken) | bot/import-bot.js, bot/bot.js |
data/bot/scenes.json, data/bot/trajectories-<tag>.jsonl |
| Reddit megathreads | not collected yet | — |
sources/update_sources.py runs every day from .github/workflows/update-sources.yml. It downloads the sheets
and the PDF (links in data/sources.json), rebuilds the data if anything changed, and opens a pull request that
merges itself when CI passes. A broken link opens an issue.
The bot importer only ever writes fingerprints, short labels, choice text and outcomes. It refuses anything that
looks like game prose or an account identifier. What the bot should log is in docs/BOT_DATA_PROTOCOL.md.