edtf-normalize

Deterministic prose-date → EDTF normalizer: "1980s" → 198X, "circa 1920" → 1920~, "около 1920 г." → 1920~ — instant, offline, the same answer on every keystroke. no_std + alloc, no third-party dependencies (only edtf-core).

Built for the human input boundary: values are constructed through edtf-core's model and rendered by its canonical Display, then re-parsed, so every output is valid, canonical EDTF by construction.

The return type is honest — Normalized { edtf, value, notes }, Ambiguous { interpretations }, or NoMatch — never a silent guess:

use edtf_normalize::{normalize, Outcome};

let Outcome::Normalized(n) = normalize("early 19th century") else { panic!() };
assert_eq!(n.edtf, "1801/1830");

// "12/04/1985" is DD/MM or MM/DD — the form gets both readings, not a guess.
assert!(matches!(normalize("12/04/1985"), Outcome::Ambiguous(_)));

// Freeform prose is the caller's problem, by design.
assert!(matches!(normalize("printed sometime after the war"), Outcome::NoMatch));

Pattern tables are per-language (English and Russian today); adding a locale means adding one table struct, not touching the grammar. Every judgement call — century arithmetic, BC year-zero handling, season codes, ambiguity policy — is a numbered N-decision in docs/normalize-notes.md, cited from the Note values attached to every result. Disagree with one? File an issue against the N-number.

Exposed to JavaScript through edtf-wasm's normalize().