[pattonw]

Why

The lint job's rule set was whatever the current ruff release considered default. ruff 0.16 widened those defaults substantially, so the unpinned job now reports 57 findings on code that was clean in February. The test workflow ran five interpreters at a single resolution, and the workflows were split across four files with older action majors.

What changes

• tests.yaml replaces pytest.yaml: triggers on pull_request, pushes to main, and manual dispatch. Matrix is the requires-python floor and newest (3.10, 3.14) × highest/lowest-direct. setup-uv@v6 pins the interpreter; uv sync then pytest --cov=src with term-missing and xml reports.
• lint.yaml replaces ruff.yaml and ty.yaml: ruff check ., ruff format --check . and ty check, each on pull requests.
• publish.yaml: setup-uv@v6; runs on v* tags only. The PYPI_API_TOKEN secret is still used; id-token: write is in place for trusted publishing once a publisher is registered on PyPI.
• pyproject.toml: [tool.ruff.lint] select = ["E4", "E7", "E9", "F"] pins the rule set to ruff's pre-0.16 defaults, so the job checks the same things it did before and stops drifting with releases. extend-exclude = ["*.md"] keeps ruff format (which formats Python fences in Markdown since 0.16) from collapsing the README's hand-aligned result comments.
• README badges follow the renamed workflows.

Verified locally

• ruff check . and ruff format --check .: clean with ruff 0.17.3.
• pytest on 3.10 / lowest-direct and 3.14 / highest: 15 passed, 1 xfailed each.
• ty check: 4 diagnostics (tests/test_coordinate.py 49, 53, 55: tuple arithmetic on Coordinate that the tests exercise on purpose; tests/test_questions.py 41). The last ty run in February was green with an older ty; this is tool drift, not a change here, so the ty job will be red until those four are annotated.

🤖 Generated with Claude Code