Validate realistic public pipeline output

The test suite includes programmatic builders that model public CellProfiler/CytoTable single-cell, pycytominer well, and JUMP treatment shapes (tests/fixtures/realistic.py). They are regression fixtures, not committed .h5ad files.

from cp_anndata_validator import validate
from tests.fixtures.realistic import make_pycytominer_well_profile

adata, path = ...  # write the fixture AnnData to a temporary .h5ad in your script
report = validate(path, schema="generic-cell-painting", profile_level="well")
print(sorted({i.code for i in report.issues}))

Pinned bare-pipeline baselines live in tests/test_realistic_baselines.py. How to read governance warnings versus likely false positives is covered in Expected findings and false positives.

For genuine Gallery plates (LINCS / JUMP), keep downloads outside the repo and follow Public-data pilots. Do not commit downloaded profiles or generated pilot .h5ad files.