Every deterministic component — intake, each pipeline stage, and every statistic — is tested with
pytest. The defining feature of this suite is that the answers are known: fixtures/ holds
synthetic report sets with planted hotspots, so the pipeline and the spatial statistics can be
validated against ground truth rather than against themselves.
- Known-answer fixtures. Recovering the planted hotspots (and only those) is the pass condition for the hotspot statistics. Interval-coverage checks validate the confidence intervals.
- No real data, ever. Fixtures are synthetic. No precise raw report is committed (hard rule #4);
data/raw/is gitignored. SeeCONTRIBUTING.md. - Run them.
make test(orpytest). The full merge gate ismake verify.
Determinism is a requirement: seeded pipelines and analyses produce the same dataset and figures every
run, which is what makes make reproduce a meaningful proof.