[lbougan]

Why

Which upstream source supplied a row was recorded only as free text (the sector description and the citation sentence in comment). That is the BAFU ToU citation, not a machine-readable tag: a consumer could not select rows by source, and sentier-brightway concatenated every folder into a database named bafu-2026. That is correct only while the inventory is 100 % BAFU.

Sector-based storage stays (decision of 2026-08-07, no source-named folders or files). This adds per-row provenance so folders can mix sources and consumers can select by it.

What (schema 0.2.0)

• processes.parquet: source (required, lower-kebab <publisher>-<edition>, e.g. bafu-2026) and source_version (optional, e.g. v1), placed after process_type. Exchanges inherit provenance through process_id, so exchange.yaml only gains a header note.
• metadata.json: required sources list (pattern-checked, unique, non-empty); schema_version 0.2.0.
• scripts/validate.py: the distinct processes.source values must equal sources, both directions. This makes metadata.json a trustworthy index, so a consumer can pick folders without opening parquet.
• data/: all 11 sectors regenerated by sentier-importers (run_bafu_local.py at sentier-importers 732ba80) with source = bafu-2026, source_version = v1. Every pre-existing column and row is unchanged; the exchanges.parquet files are byte-identical to main (same hashes in sentier-brightway's pin file).
• README and data/README.md: provenance rule rewritten, source-id convention documented.

Verification

$ python scripts/validate.py
OK — 11 sector folder(s) validate against the schema contracts.

With a throwaway folder whose sources says other-1 but whose parquet says bafu-2026, the validator reports both directions:

- 98-throwaway/processes.parquet: source values not declared in metadata.json sources: ['bafu-2026']
- 98-throwaway/metadata.json: sources declared but absent from processes.parquet: ['other-1']

Compatibility

Breaking for a reader that rejects unknown columns or requires the 0.1.0 metadata shape. Known reader: sentier-brightway, pinned by commit and updated in its own feat/source-column branch. Please merge without squash so the pinned commit 2837e1b stays reachable; otherwise the brightway PR re-pins to the squash commit.

Landing order

1. sentier-dev/sentier-importers feat/source-column (emits the columns)
2. this PR
3. sentier-dev/sentier-brightway feat/source-column