tests/test_<module>.py mirrors src/tn_noise_learning/<module>.py; test_end_to_end.py recovers a known channel
from simulated shots.
Fixtures are tiny: Topology.chain(2) and Topology.square(2, 2). Dense references are built with explicit Kraus
sums and qiskit.quantum_info.Operator, never with another tensor-network path.
Use pytest.raises(..., match=...). Mark tests that take more than about 10 s with @pytest.mark.slow.
Warnings are errors. A third-party warning that cannot be fixed gets a narrowly scoped ignore in
[tool.pytest.ini_options], with a comment saying why.
Pass a seed and use np.random.default_rng; never seed global state.
CI runs the suite on Linux x86-64, Linux ARM64 and macOS with Python 3.12 and 3.15 (.github/workflows/test.yml),
and once with coverage (qa-analysis.yml).