RELEASE_CHECKLIST_v0.1.0

1. Preconditions

  • All ambiguous words removed from CRF golden set
  • All tests passing locally (make test)
  • Google Sheet URL + sheet name confirmed
  • No model files present in src/zomi_syl/models/crf/
  • Version bumped to 0.1.1 in code (will finalize in step 8)

2. Refresh Data

  • Fetch latest human syllabified dataset
    make get-zomi-syllabified-human URL=<sheet-url> SHEET=<sheetname>
  • Clean dataset
    make clean-dataset
  • Confirm training/data/zomi_only.tsv exists and is non‑empty

3. Train CRF Model

  • Train CRF
    make train-crf
  • Confirm output files in training/model/crf/:
  • crf_syllabifier.joblib
  • config.json
  • stats.txt
  • eval.txt

4. Regenerate Golden Regression

  • Regenerate golden CRF outputs
    python scripts/get_golden_crf_frozen_data.py
  • Confirm:
  • tests/golden/crf_golden.tsv updated
  • No ambiguous words included
  • No multi‑variant words included
  • Optional: inspect CRF predictions
    make check-crf WORDS="ama upa zawlai"

5. Freeze + Package Model

  • Run release freeze script
    ./scripts/release_crf_freeze.sh --url <sheet-url> --sheet <sheetname>
  • Confirm:
  • Wheel built in dist/
  • Temporary model removed from repo
  • No model files left in src/zomi_syl/models/crf/

6. Validate Wheel

  • Install wheel
    pip install --force-reinstall dist/zomi_syl-*.whl
  • Run full test suite
    make test
  • Confirm:
  • All tests pass
  • Golden regression matches
  • CRFBackend loads packaged model

7. Documentation Updates

  • Update CHANGELOG.md for v0.1.0
  • Update README if needed:
  • Installation
  • Example usage
  • Backend notes
  • Release notes

8. Versioning

  • Update version in:
  • src/zomi_syl/version.py
  • pyproject.toml
  • Commit changes
    git commit -am "Release v0.1.0"

9. Tag + Push

  • Tag release
    git tag v0.1.0
git push --tags

10. Publish to PyPI

  • Upload
    twine upload dist/*

11. Post‑Release Validation

  • Install from PyPI in a clean environment
  • Run: 
    python -m zomi_syl syllabify "themthum"
  • Confirm correct output
  • Confirm CRFBackend loads packaged model

12. Optional Enhancements

  • Add ambiguous‑word detector
  • Add golden‑set enforcement
  • Add CRF vs rule‑based diff tool
  • Add transformer backend regression tests