Release 1.15.0

Added

• Add generated country metadata, failure-reason and input-format guides, executed and strictly typed documentation
  examples, and contributor/country workflow templates
  (#328).

• Add primary-only normalize_id(), format_id() and get_input_mask() with frozen InputMask results.
  Metadata-driven formatting checks length, not validity; input patterns use ASCII uppercase letters/digits.
  Swedish formatting preserves a significant + in both 10-digit and 12-digit forms; preferred mask patterns
  remain literal minus layouts. Greece's primary metadata example now uses already-accepted Latin letters;
  Greek validation and formatting remain unchanged (#321).

• Add failure_reason(id_class, id_number) for primary and secondary ID types. Unified validation and parsing now
  report best-effort length, format, checksum and embedded-calendar-date reasons; inconclusive checks and exceptions
  retain VALIDATION_FAILED. Date tracing is isolated between nested calls, threads and async tasks, and country
  validity, parser and checksum behavior is unchanged (#322).

• Add idnumbers.parse_id_info(country, id_number) for a country's primary NationalID, returning the frozen
  ParseSuccess / ParseFailure union ParseIdInfoResult. The ok discriminator supports strict type narrowing;
  success includes a fresh dictionary of the country parser's information. Lookup and validation failures match
  validate(), while valid IDs without extractable information give NOT_PARSABLE. Country parsers are unchanged
  (#319).

• Add a country registry, idnumbers.registry, which is re-exported from idnumbers. get_validator('tw'),
  get_country('TWN') and resolve_country('tw') find the NationalID class, a CountryEntry (alpha-3 and alpha-2
  code, name, NationalID and all ID types) or the alpha-3 code of a country by its alpha-2 or alpha-3 code in any
  letter case, and give None for anything unknown instead of raising. list_supported_countries() lists the 78
  countries, and register() adds a custom country. The names are the English names of Unicode CLDR 48.2.3. AUS and
  GRC keep their current NationalID, and a country module is imported on the first lookup of its country
  (#318).

• Add idnumbers.validate(country, id_number), which validates an ID number of any supported country and returns a
  frozen ValidationResult with is_valid, the alpha-3 country_code, the id_number, the extracted_info of a
  valid ID number (the parse() result, None when the ID type isn't parsable), a reason and an error_message. It
  never raises, and a result is truthy when it is valid. An unknown country gives the reason unsupported_country,
  other failures a best-effort diagnostic reason. validate_many() validates (country, id_number) pairs, and
  FailureReason lists the reasons, which later releases may extend. The country classes are unchanged
  (#317).

• Describe every ID class with seven more METADATA keys: country_name, id_type, official_name (None for an
  ID with no local-language name), display_format, example (a synthetic number the class validates),
  checksum_algorithm (None without a check digit) and masks, the layouts the class accepts. METADATA is now an
  IdMetadata, a typed SimpleNamespace subclass in idnumbers.nationalid.metadata, so attribute access, vars(),
  copy() and getattr() work as before. A class made by alias_of, such as AUS.NationalID, now has the name,
  module and docstring of its target instead of AliasType, which changes its repr. CAN.SocialInsuranceNumber
  gets the deprecated key the other classes already had, and tools.scan_ids lists the classes through the country
  registry (#320).

Changed

• Ship the PEP 561 py.typed marker, so type checkers use the package's annotations, and correct the annotations
  that disagreed with the runtime values: the CHECK_DIGIT alias, the untyped normalize and checksum helpers,
  the COL and KAZ checksum methods that return False for malformed input, the AUS TFN and KWT civil-number
  checksum methods, which can return 10, and the LTU extract_year_base_gender result, which is never None
  (#325).

• Move the build metadata from setup.py to a PEP 621 pyproject.toml. The license is now the SPDX expression MIT
  instead of the deprecated license classifier (#465), and
  building from source needs setuptools 77 or later, which pip and uv install automatically
  (#325).

• CI type-checks the package with strict mypy for Python 3.9 and checks the built distributions with
  twine check --strict (#325).

• Re-export the ID classes explicitly (import X as X) from the country modules such as AUS, so strict type
  checkers (mypy --strict, which disallows implicit re-exports) accept
  from idnumbers.nationalid.AUS import MedicareNumber and AUS.MedicareNumber; nothing changes at runtime
  (#326).

• Prepare isolated, guarded PyPI publication and opt-in Trusted Publishing with attestations; retain the working
  default token route and existing secrets. External publisher setup and the first verified trusted release remain
  pending maintainer action. Document release ordering and 1.14.0-to-1.15.0 migration
  (#327).

• Add dev-only Hypothesis checks of arbitrary Unicode inputs for every registered ID type, metadata method
  contracts, out-of-range lengths and checksum mutations, with explicit synthetic witnesses for checksum limits
  (#324).

• CI lints with ruff, enforces a 97% line-and-branch coverage floor, runs the README Python examples and
  informational benchmarks, keeps GitHub Actions updated with Dependabot, and reports a single ci-summary result
  (#326). Formatting with ruff format and a coverage badge
  await a decision in #485.

• Vendor the Node port's parity corpus (78 countries, 17,960 vectors) at a pinned port commit and record Python's own
  validity for it, so a validity change on any shared vector fails a test and gets a port note;
  scripts/sync_parity_corpus.py refreshes both (#323).