Source edibility for real: Kew useful-plants checklist, PFAF, GRIN — nothing fills these fields today

[lambda2]

Part of #350. Feeds #354, #355 and #356.

Nothing currently sources edibility. species_facts holds 3 edible_part rows (source pfaf) and 71 vegetable rows (source merged_from_synonym, i.e. copied from a synonym, not sourced at all). There is no PFAF crawler in lib/crawlers/; those 3 rows are residue.

Ranked by coverage-per-effort:

1. Kew — World Checklist of Useful Plant Species

~40,000 species with use categories including Human Food, published open with a DOI. By far the best ratio available, and the decisive advantage: it is keyed on WCVP accepted names, which we already ingest (lib/crawlers/wcvp.rb). Name reconciliation — the part that usually eats the budget on a job like this — is close to free here.

Gives a coarse boolean-grade signal (this species is used as human food), not parts. That is the coverage layer.

2. Plants For A Future

~7,000 species with an Edibility Rating 0-5, the edible part, and preparation notes — precisely the three columns #354 defines, which is why the rating scale in that issue is PFAF's rather than one of ours. This is the depth layer: fewer species, but the only source that fills preparation and caution at scale.

Licence needs settling before ingest — PFAF's database terms are not a standard open licence. Check, and if it does not permit redistribution, do not ingest it: cite it as a link-out instead. Same discipline as every source under #318.

3. USDA GRIN Taxonomy

Economic-use records for cultivated and crop-adjacent species. US-government public domain, stable identifiers, good API. Strongest on the cultivated side of #358.

4. OpenFarm

Already wired (lib/crawlers/openfarm.rb). Growing guides for crops — feeds cultivated and the planting_* fields, not edibility per se. Cheap because the crawler exists.

5. USDA PLANTS palatable_human

old_species carries the column from the original import. It is empty in the dev restore (0 true, 0 false, 0 null — the table itself is empty there), so verify against prod before counting on it. If prod has values it is a few thousand US species for free; if not, drop this line.

Scope

• Crawlers land in treflehq/trefle-crawlers (private) and write SpeciesFact rows with source, source_record_id, source_url, evidence_type: reported. Same contract as the TRY import (Promote TRY trait facts into empty species columns, with attribution #328).
• This issue is the public-side half: a ForeignSource row per source with its licence and citation text (Expose each upstream source's licence in the API (sources[].licence) #318), the promotion migrator that projects the strongest active fact onto the columns and the Replace the edible_part bitmask with sourced part rows: preparation, caution, and a 0-5 food-use rating #354/Toxicity is an enum filled on 0 of 489k species: populate it, and model it per part like edibility #355 tables, and a lib/checks/ divergence report where two sources disagree on the same species.
• Never overwrite a filled value. Promote into empty fields only; disagreements go to the corrections queue for a human. Identical rule to Promote TRY trait facts into empty species columns, with attribution #328, and it matters more here: a source that wrongly asserts edibility is worse than a gap.
• Licence recorded before ingest, no exceptions. An edibility dataset we cannot legally redistribute is a liability, not an asset.

Acceptance

• Per-source: ForeignSource with licence + citation, a fact count, and a sample of 20 records verified by hand against the source.
• Promotion migrator specs: fills empty only, idempotent, skips facts that fail Checks::PlausibleValues.
• Divergence check reports conflicts without writing.
• Coverage report posted here — species with an edibility fact, before and after each source.
• Attribution flows to the citation page (/citation page, CITATION.cff and the /about licence section #321) and per-source licence exposure (Expose each upstream source's licence in the API (sources[].licence) #318).

Touches: treflehq/trefle-crawlers (private), app/workers/migrators/, lib/checks/, db/seeds, spec/