The tracked data source is split into three parts:

• canonical/ contains the normalized JSONL used to build the database.
• migrations/ contains forward-only SQLite schema changes.
• patches/ contains small, sourced content changes.

canonical/ is generated by npm run data:canonical:export. Do not edit it by hand. A factual correction belongs in a new patch; a schema change belongs in a migration.

canonical data
  + migrations
  + patches
  -> .cache/equilibrium.sqlite
  -> .generated/documents/

Run the full rebuild with:

npm run data:rebuild

The SQLite database, generated shards, and reports are build outputs. They are ignored by Git and should not be copied into public/.

The old compressed seed is no longer part of the working tree. Its source documents and original records are retained under canonical/provenance/; the seed itself remains available in Git history at commit 43c23873.

Start by finding and inspecting the existing record:

npm run data:find -- --query "Seismic wand"
npm run data:context -- --id item:seismic-wand --format markdown
npm run data:impact -- --id item:seismic-wand

Then:

1. Add a small JSONL patch under patches/.
2. Include stable IDs and source links.
3. Run npm run data:rebuild.
4. Run npm run data:canonical:export.
5. Run npm run data:canonical:validate and npm run audit:data.
6. Commit the patch and the matching canonical export together.

Applied patches are immutable. A later correction should be another patch, not an edit to the old one.

• Do not hand-edit canonical/, .cache/, or .generated/.
• Do not rewrite a whole dataset to change one fact.
• Use migrations for schema changes only.
• Keep unknown or unrevealed league values empty until they have a source.
• Do not restore the retired per-domain JSON exports or mutation scripts.
• Do not create a separate Troll Country region. It is part of Asgarnia in this project.

The full workflow and patch formats are documented in docs/data-platform.md.