kert 35dc05c4d0 feat(nlm): UMLS / SNOMED CT / RxNorm releases and VSAC value sets ...

A UMLS licence key (UMLS_API_KEY, read via conf.secret("nlm.api_key"))
now drives a new `nlm` module and `stack nlm` CLI:

- `stack nlm releases` lists the current file of every UTS release
  stream (14 today: RxNorm full/prescribable/weekly + RxNav-in-a-Box,
  SNOMED CT US/International/Spanish/CORE/transitive closure/ICD-10-CM
  map, UMLS full/Metathesaurus/MRCONSO).
- `stack nlm fetch [--stream …]` downloads them through the UTS
  /download endpoint to `<nlm.dir>/<stream>/<version>/<file>`, with a
  sidecar (what, where from, bytes, sha256, when). A finished file is
  skipped; an interrupted one resumes from its .part with a Range
  request; a non-zip answer (a login page) is refused.
- `stack nlm vsac` saves every VSAC value-set definition (18.6k
  versions, paged FHIR search) as gzipped JSON plus index.jsonl, then
  $expands each one — large expansions are paged and joined — with
  backoff on 429/5xx, 4 workers, resumable, failures to errors.jsonl.

The key rides as a query parameter (UTS) or basic auth (VSAC);
nlm.scrub() strips it from every error before it can reach a log.
nlm.dir defaults to the gitignored data/nlm: the content is
UMLS-licensed and must never be published.

Live-checked: listing, an authenticated download with sidecar, and a
768-code value set expanded in pages of 25 matching the one-shot
expansion exactly. Tests mock httpx (100% of the new module).

2026-10-27 13:00:28 -04:00

1.0 MiB

Raw Blame History

View Raw