@discretion/data

Generated data tables for @discretion/core.

You almost certainly do not want to install this directly. It is a dependency of @discretion/core, published separately only so that core can pin it at an exact version. Install core.

What is in it

| export | what it is | | --- | --- | | CONFUSABLES | Unicode confusable-character mappings, used by Stage 0 homoglyph folding | | TRIGGER_LEXICONS | Per-language context trigger terms for Stage 3 scoring | | DOMAIN_LEXICONS | Vocabulary used to profile a document's domain | | GAZETTEERS | Name, organisation and place lists for Stage 2b | | SECRET_PROVIDERS | API-key prefix table — the reason a new provider needs no code change | | SEVERITY_WEIGHTS, CATEGORY_OF, TYPE_FACTORS | Exposure scoring inputs | | CALIBRATION_MODEL | The fitted confidence calibration curves |

These are generated, and their values are not stable

Every table here is produced by a generator, from a Unicode release, a corpus, or an evaluation run. A value can change in a patch release: a confusables entry, a gazetteer name, or a calibration curve refitted after a corpus change.

The package version tracks the type surface, which follows VERSIONING.md. Do not build anything that depends on a specific numeric value in CALIBRATION_MODEL or a specific entry in CONFUSABLES surviving an upgrade.

Regenerating requires network access (npm run generate). The generated modules are committed so that building the project does not.

Licence

MIT, except the Unicode-derived data, which is under the Unicode licence. Attribution and terms are in THIRD_PARTY_NOTICES.md in this package.