discrimination/T2_examples.md

STEM BIO-AI Stage 3 Discrimination Examples: T2 Domain Tests

T2 awards +15 for DOMAIN-SPECIFIC regression tests. This means tests that verify biological, chemical, or clinical correctness of outputs -- not infrastructure tests.

---

QUALIFIES as T2 (+15 pts)

• test_scrnaseq_clustering.py -- tests specific input produces expected cell type clusters
• test_gwas_pipeline.py -- tests known causal variants correctly identified
• test_docking_output.py -- tests AutoDock Vina binding affinity within expected range
• test_pharmacogenomics.py -- tests CYP2D6 *4 allele triggers correct CPIC recommendation
• test_variant_classification.py -- tests known pathogenic ClinVar variant correctly classified
• test_smiles_validity.py -- tests generated SMILES parse correctly in RDKit with valid properties

DOES NOT QUALIFY as T2 (score at T1 level or +0)

• test_governance_boundaries.py -- tests gate mechanics, not biological accuracy
• test_api_integration.py -- tests external API calls succeed (infrastructure)
• test_file_format.py -- tests output file structure (format, not accuracy)
• test_skill_validation.py -- tests SKILL.md sections present (structural)
• pytest with only assert result is not None -- null check, not domain verification

BOUNDARY CASE

• test_input_validation.py that checks scRNA-seq h5ad format: T2 PARTIAL -- tests domain data format but not output accuracy. Award +8 (T1 coverage-unstated level), not +15.

CODE_PATH Verification (LOCAL_ANALYSIS)

# Find domain-specific test files
find . -name "test_*.py" -o -name "*_test.py" | \
  xargs grep -l "CYP\|CPIC\|allele\|SMILES\|binding_affinity\|cluster\|GWAS\|variant"

# Distinguish from infrastructure tests
find . -name "test_*.py" | \
  xargs grep -l "governance\|api_key\|file_format\|schema\|connection"