KazStem 0.2.1

KazStem 0.2.1

KazStem 0.2.1 is a packaging-only patch over 0.2.0. The analyzer,
generator, resource contract, and published evaluation results are unchanged.

What changed:

• removed an invalid PyPI classifier that blocked the first trusted-publishing
  attempt (PyPI accepted no 0.2.0 artifact);
• added checksum plus embedded package-name/version validation before trusted
  publishing;
• published the first installable kazstem distribution on PyPI;
• added a self-contained Ubuntu 24.04 x86-64 archive with the sealed Kazakh
  resources and native HFST/CG runtime.

Install the Python package:

python -m pip install kazstem==0.2.1

The py3-none-any wheel is the Python front end. Building resources from
source currently follows the Ubuntu 24.04 x86-64 bootstrap documented in the
README. For a ready-to-run Linux download, use
kazstem-0.2.1-linux-x86_64-ubuntu24.04.tar.gz; it is specifically labeled for
Ubuntu 24.04 / glibc 2.39 and must not be treated as a generic Linux binary.

There are no placeholder macOS or Windows executables. Portable native builds
are tracked in issue #1 and
will be attached only after platform-native HFST/CG dependency, licensing, and
black-box behavior gates pass. Neural model weights are not redistributed.

Validation uses the same frozen 0.2 behavior baseline: 323 tests, 322 passing,
one intentionally documented CoNLL-U MWT skip, and zero failures. The Ubuntu
archive additionally passed clean-extraction analysis, OOV, generation,
Constraint Grammar, JSON/XML, dependency-closure, and redistribution-notice
checks.

See SHA256SUMS for artifact identities, THIRD_PARTY.md for licensing, and
docs/MYSTEM_COMPATIBILITY.md for exact MyStem compatibility boundaries.

KazStem 0.2.0

Superseded by KazStem 0.2.1. PyPI rejected 0.2.0 because its metadata
contained an invalid Trove classifier; PyPI accepted no 0.2.0 artifact. The
public 0.2.0 tag and assets remain immutable for reproducibility.

KazStem 0.2.0

This is the first public KazStem source release: an ambiguity-preserving Kazakh
morphological analyzer and generator with a MyStem-like CLI.

Highlights:

• canonical kazstem command, with qazmorph and mystem-kz retained as aliases;
• complete finite-state candidate lattices and exact raw/morpheme provenance;
• bounded productive OOV analysis for eligible Kazakh noun, adjective, and verb stems;
• candidate-constrained Constraint Grammar and optional neural ranking;
• text, MyStem-shaped JSON/XML, lossless JSONL v2, and CoNLL-U output;
• content-addressed resource/toolchain manifests and post-run verification;
• CFF, BibTeX, CodeMeta, and Zenodo-ready citation metadata.

Validation: 323 tests ran against the sealed f03e resource bundle on the
reference host; 322 passed and one explicitly documented CoNLL-U MWT limitation
was skipped. Citation metadata validates against CFF 1.2.0. A fresh offline
wheel install exposed all three console commands and completed a real Kazakh
analysis smoke test.

The py3-none-any wheel is the Python front end. It still needs the separately
verified Kazakh FST/CG resource and native runtime described in the README; it
is not a standalone macOS or Windows executable. Neural model weights are not
redistributed.

See SHA256SUMS for artifact hashes, THIRD_PARTY.md for licensing, and
docs/MYSTEM_COMPATIBILITY.md for exact MyStem compatibility boundaries.