Testing¶

Unit Tests¶

uv run pytest                          # all tests, with coverage
uv run pytest tests/py_tests/test_io.py
uv run pytest -m "not slow"            # skip slow tests
uv run pytest -p randomly              # explicit randomised order

Pytest is configured in pyproject.toml with --cov=bsllmner2 --cov-report=term-missing, strict markers, and async support. Property-based tests are written with hypothesis.

Markers¶

Marker	Description
`slow`	Long-running tests. Skip with `-m "not slow"`.

Type Checking¶

uv run mypy

mypy runs in strict mode with the pydantic.mypy plugin against bsllmner2/**/*.py and tests/**/*.py. External modules without stubs have their missing-imports ignored via [[tool.mypy.overrides]] in pyproject.toml.

Linting and Formatting¶

uv run ruff check bsllmner2/ tests/ scripts/
uv run ruff format bsllmner2/ tests/ scripts/
uv run ruff format --check bsllmner2/ tests/ scripts/

Ruff is configured with select = ["ALL"] and an explicit ignore list (see [tool.ruff.lint] in pyproject.toml). target-version = "py310" and line-length = 120.

Mutation Testing¶

mutmut validates that the test suite detects code mutations.

uv run mutmut run
uv run mutmut results

Configured in pyproject.toml:

[tool.mutmut]
paths_to_mutate = ["bsllmner2/"]
tests_dir = ["tests/py_tests/"]
pytest_add_cli_args = ["--no-cov", "-p", "no:randomly"]

Model Evaluation¶

tests/model-evaluation/ benchmarks LLM models on the ontology mapping task using the 600-entry evaluation set in tests/data/eval_biosample.json with tests/data/eval_gold_standard.tsv as the gold standard for the cell_line field.

Source datasets (Zenodo):

Metrics produced: accuracy, precision, recall, F1 (see SelectResult.evaluation in Data Formats). For how to interpret performance numbers in the result JSON, see Benchmarking.

Full workflow (batch execution across models, metric aggregation): see tests/model-evaluation/README.md.