sparam-conformance / CONTRIBUTING.md
nickh007's picture
Initial release: 11 labelled networks with verified ground truth, plus a scorer
b10c2aa verified
|
Raw
History Blame Contribute Delete
1.49 kB

Contributing to sparam-conformance

The most useful contribution is a new pathological case

If you have an S-parameter file that a checker got wrong — passed something non-physical, or rejected something real — that is the highest-value thing you can send.

Every case must be synthesised, not captured

Add a closed-form generator to generate.py. The label must be derivable from the construction: we know a network is non-passive because we built gain into it, not because a tool said so. Captured measurement data has no ground truth, which is exactly the problem this corpus exists to fix.

Every label is re-derived in the tests

tests/test_conformance.py recomputes passivity, reciprocity and energy for every case by independent linear algebra and asserts they match the manifest. Add your case and the tests will check your label. If they disagree with you, they are probably right — that check caught three wrong labels during initial development.

Isolate one law where physics permits

A case that breaks four laws at once tells you nothing about which check is alive. Where a single-law violation is impossible (a reflection coefficient above unity necessarily breaks energy conservation too), say so in the note and label all the affected laws honestly.

Determinism

Generation must be reproducible byte-for-byte; there is a test. No randomness without a pinned seed.

python generate.py
PYTHONPATH=.:../sparam-lint/src pytest tests/ -q