Feature request: traits extracted from Wikipedia introductions via taxoniq, as a new source #295

Proposal: a new programmatic source built from the Wikipedia introductory paragraphs that Levi's taxoniq release carries for every taxon with a Taxonbar article (lwaldron/taxoniq, built 2026-10-31; Taxon.description, Taxon.wikidata_id). Traits would be extracted with the same prompt-based approach and validation as bergeys_2026, and the source would live in bugphyzzExports sources/.

Measured against the devel export of 2026-11-10:

• 10,149 of the 232,140 records are Bacteria or Archaea: 7,643 species, 1,951 genera, about 500 above genus.
• Paragraphs that mention a trait: gram stain 5,944; aerophilicity 3,467; motility 2,703; habitat 2,434; shape 2,322; spore formation 1,457; pathogenicity 512.
• Only 708 of those taxa are absent from bugphyzz. Taxa whose article mentions an attribute they have no curated value for: gram stain 2,879; aerophilicity 3,522; motility 4,616; shape 4,255; spore formation 6,198. Upper bounds, before extraction and conflict resolution.

So the gain is a few thousand annotations against 366,000 curated ones, concentrated in attributes that are already well covered. The more interesting uses are the 708 uncovered taxa and attributes bugphyzz has no source for yet (habitat terms, pathogenicity).

Before building:

1. Licence. The extracts are CC BY-SA 3.0; bugphyzz is CC BY 4.0. Decide whether derived trait rows carry the share-alike condition.
2. Evidence and confidence: tertiary text, so NAS and low, like BacMap.
3. Pin the taxoniq release the extracts come from, as ncbi_assemblies/python.sh pins the taxon database.

Not for v1.1.0. Filed so the measurement is not lost.

Drafted with Claude Code.