[tjkessler]

Release prep for 0.16.0. #65 (mass-action reaction-network datasets) is fully on main. This PR changes only versions and release metadata; there are no code changes besides __version__.

Changes

• src/koopman_graph/__init__.py: __version__ = "0.16.0".
• CHANGELOG.md:
  • [Unreleased] → ## [0.16.0] - 2026-10-15, with a summary paragraph like 0.15.0's: reaction-network datasets with oracle references, optional [reaction] / [cantera] extras, defaults and FORMAT_VERSION unchanged. An empty [Unreleased] stays on top.
  • The skeleton-era entry from Skeleton for koopman_graph.datasets.reaction (#65) #66 said nothing was exported from koopman_graph.datasets, which stopped being true when ReactionNetworkBenchmark with packaged FAIR cards (#65) #74 exported ReactionNetworkBenchmark. It now describes the released state.
  • Adds the [0.16.0] compare link (0.15.0...0.16.0).
• CITATION.cff: version: 0.16.0, date-released: 2026-10-15. The DOI stays 0.15.0's until Zenodo mints 0.16.0's, following the 0.15.0 precedent (Release KoopmanGraph 0.15.0 (identification and evidence surfaces) #56, then Record Zenodo DOI for the v0.15.0 software archive #57). test_release_metadata_versions_agree pins that DOI.
• README "What's new in 0.16.0" and BibTeX version, paper.bib software version, the docs index.rst version paragraph, the installation.rst pin, and SECURITY's supported line all move to 0.16.0. The version-consistency tests require these to agree with __version__.

Left for the Zenodo follow-up: the version DOI in CITATION.cff, README BibTeX, paper.bib, the paper.md text, the README DOI badge, and the test_docs.py DOI pin.

Release checks run locally (mirroring release.yml)

• verify job, with the same uv sync --frozen extras (now including reaction and cantera):
  • ruff check and ruff format --check src/ tests/ pass.
  • pytest tests/ -n auto --cov=koopman_graph --cov-fail-under=90 (full suite, no marker filter): 3560 passed, 15 skipped, 97.35% coverage. The run first showed 1 failure, test_installed_version_matches_metadata. That was a stale local editable install still reporting 0.15.0 metadata, because uv sync doesn't rebuild an editable package when only its dynamic version changes. After uv sync --reinstall-package koopman-graph it passes, along with all tests/package. CI's fresh sync isn't affected.
  • sphinx-build -b html -W --keep-going passes.
• pypi-publish build: python -m build produces koopman_graph-0.16.0-py3-none-any.whl and koopman_graph-0.16.0.tar.gz, and both pass twine check. The wheel metadata has Version: 0.16.0 plus Provides-Extra: reaction and cantera, it contains all 6 reaction cards, and it includes no tests or examples.
• Install check against the built wheel, since PyPI doesn't have it yet. In a fresh Python 3.12 venv run with python -I:
  • koopman_graph.__version__ == "0.16.0", all six cards are listed, and reaction_chain_v1 generates;
  • importing loads neither SciPy nor Cantera, and the stiff path's error points at pip install koopman-graph[reaction].
• uv lock --check passes. The project's dynamic version isn't recorded in uv.lock.

After merge

1. Create the GitHub Release with tag 0.16.0 (earlier tags have no v prefix). Publishing it triggers release.yml, which runs verify and then pypi-publish. Approve the pypi environment when prompted. Zenodo's webhook archives the tag and mints the version DOI.
2. If the release goes out on a different day, update date-released and the CHANGELOG date to match (0.15.0 was dated 2026-08-19 but published on 08-21).
3. pip install koopman-graph==0.16.0 and check __version__.
4. Open the DOI follow-up PR, as Record Zenodo DOI for the v0.15.0 software archive #57 did.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
src/koopman_graph/__init__.py	100.00% <100.00%> (ø)

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.