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
What files and fingerprints does a run produce?	Outputs
How do I compare completed run reports?	Run Comparison
Can this installation run the shipped example workflows?	Integration Tests
Do repeated runs match across machines or environments?	Cross-Environment Reproducibility
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 WES/WGS output reproducibility, prioritize the final VCF calls fingerprint, then inspect the stricter full-record VCF fingerprint. File inventories and logs are useful audit evidence, but they can differ between environments even when final variant calls match.