nl2time (Python)

Deterministic resolution of temporal-expression IR — the Python port of nl2time, a bidirectional natural language ⇄ date/time library built around a JSON intermediate representation (TimeExpr).

Scope of this release: the language-neutral core. IR validation, TimeContext (reference instant, IANA timezone, locale-derived week/date policy, ambiguity knobs), and the full deterministic engine resolve() — every operator including calendar snapping, DST-correct offsets, holidays (computed Easter), business-day spans, and ordered ambiguity candidates. Natural-language parsing and rendering are not yet ported; today the expected producers of IR are the JS parser, an LLM emitting schema-constrained IR, or your own code. Time model: whenever (Temporal-inspired).

from nl2time import TimeContext, resolve

ctx = TimeContext.make({
    "now": "2026-07-20T17:00:00Z",
    "timeZone": "America/Los_Angeles",
    "locale": "en-US",
})

# "last week" as IR (e.g. produced by the JS parser or an LLM):
expr = {
    "op": "snap",
    "base": {"op": "offset", "base": {"op": "now"}, "amount": -1, "unit": "week"},
    "unit": "week",
}
resolve(expr, ctx)["candidates"][0]
# {"kind": "interval", "start": Instant("2026-07-12 07:00:00Z"),
#  "end": Instant("2026-07-19 07:00:00Z"), "grain": "week"}
# (values are `whenever.Instant`; candidates are ordered, ambiguity is data)

Parity guarantee

This engine reproduces the JS reference bit-exactly — instants, grains, and candidate order — across all 2,782 machine-generated engine-parity fixtures derived from the project's six-language conformance corpora, enforced in CI on every commit. Spec: ir-spec.md; strategy: porting.md; test-data provenance: corpus/ATTRIBUTIONS.md.

MIT.