Validation & Reproducibility

CBIcall separates framework validation from biological validation of the underlying variant-calling methods. This section collects the evidence and tools used to audit CBIcall executions.

Question	Page
Can this installation run the shipped example workflows?	Integration Tests
Do repeated runs match across machines or environments?	Cross-Environment Reproducibility
How do I compare completed run reports?	Run Comparison
Are installed resource bundles compatible with selected workflows?	Resource Validation
How do native WES calls compare with a truth set?	GIAB Benchmarking

Scope

Integration tests and run comparison validate the CBIcall execution contract: parameters, workflow resolution, resource identity, runtime evidence, and output fingerprints. They do not replace biological or clinical validation of GATK, MToolBox, nf-core/Sarek, or other upstream methods.

Evidence Layers

CBIcall records reproducibility evidence at several layers:

• log.json records resolved parameters, runtime profile, workflow selection, and resource identity.
• cbicall-execution-contract.json records the backend-ready launch plan.
• run-report.json records workflow fingerprints, resource fingerprints, output inventories, final-output hashes, software/runtime evidence, and status.
• cbicall compare-runs compares completed reports across independent runs.

For output reproducibility, prioritize normalized final-output fingerprints. File inventories and logs are useful audit evidence, but they can differ between environments even when final VCF records match.