Quality gate

Run this before declaring any task complete. All Python targets go through uv; the project declares Python 3.13 in pyproject.toml.

Commands

make lint runs, in order: ruff format --check ., ruff check ., uv run mypy, uv run basedpyright. Run it exactly as CI does rather than invoking the tools individually, so a green local run means a green CI run.

What the gate enforces

• ruff: line length 88, target py313, rules E4, E7, E9, F, I, UP, B, SIM, RUF. Auto-fixable findings are handled by make format.
• mypy (src and tests, mypy_path = ["src"]) and basedpyright (typeCheckingMode = "standard"). The project holds zero warnings as debt: src/, tests/, and the eval notebook all check clean. Do not silence a finding to get green; fix the cause.
• pytest: testpaths = ["tests"], with -m 'not eval' so make test stays a fast unit run. The eval marker is the notebook suite.

Test surface

• Tool tests exercise raw callables captured by the FakeMCP fixtures in tests/conftest.py.
• tests/test_tool_schemas.py asserts the wire contract from tools/list.
• tests/test_smoke_stdio.py launches the real server as a subprocess and drives a JSON-RPC session; it is the guard for the "never write to stdout" rule and for the exact registered tool set.
• tests/test_eval_notebook.py runs notebooks/eval_msmcp.ipynb under the eval marker.
• tests/test_spearman.py guards the tie-aware correlation reported by the similarity validation (it runs in the fast suite; no optional dependency).
• tests/test_validation_harness.py covers the shipped validation harness's protocol — the query/database split, the leakage guard, and the absent-versus-invalid dataset behaviour — with neither matchms nor the dataset.
• validation/ holds the opt-in scientific-validation harness. It is not collected by pytest and not part of the gate; matchms is supplied only for make validate-similarity and is never a runtime dependency.

Conventions the gate assumes

• Prefer typed interfaces; every public callable is annotated.
• Preserve public APIs unless the task explicitly changes them.
• Add tests for behavioural changes.
• Keep changes focused; do not fix unrelated failures.
• drafts/ is gitignored scratch and excluded from lint and type checking; nothing shipped may live there.
• validation/ is shipped and type-checked (it is in mypy's files); it must import cleanly without matchms, so the fast gate passes on a machine that never installs the optional validation dependency.

Checklist

• make test passes (or the specific new test does).
• make lint passes with no new suppressions.
• Behavioural changes have tests.
• If the notebook changed, make eval passes.