Data licenses
The Apache-2.0 license covers the code. It does not cover the data the code reads, which comes from public sources that each carry their own terms.DATA_LICENSES.md in the package is the
record; this page summarizes it.
Bundled base-rate tables
The singling-out and clinical-rarity controls read five derived aggregate tables undersrc/tacit_deid/base_rates/data/. They are proportions and counts, not microdata.
Every row of
national-prevalence.json names the page its figure was read from and the licence of
that page. The package’s test suite admits only licences that allow commercial redistribution and
recomputes each proportion from the published value. Definitions differ between countries (for
example, the US hypertension figure uses a 130/80 threshold and the Australian one 140/90) and are
recorded on each row. Where no openly licensed page stated a figure, the row is absent.
The provider exposes demographic joint proportions (US only) and single-condition prevalence
marginals. It has no table of clinical joint distributions, because those are not published; that
is a stated boundary condition of the engine, not a gap to patch.
Detector models, downloaded on demand
No GLiNER model is bundled. Thegliner extra is part of the default install, and the weights
download from Hugging Face on first use, or load from a local directory named by
TACIT_DEID_GLINER_PII_OPEN_MODEL, TACIT_DEID_GLINER_PII_MODEL, or TACIT_DEID_GLINER_MODEL.
Datasets behind the measurements
The numbers on the Verification page were taken on synthetic datasets that Tacit holds for evaluation. None of them is inside the package.
The Privasis-Cleaner model is non-commercial and was used for research comparison only; it is not
part of the SDK.