Pipelines can lose real variants. WAVE checks the evidence.
A genomic workflow turns read evidence into a final call list. A filter can remove a real variant, and a caller can never propose one. WAVE checks each pipeline decision against the reads and searches the reads for what no call list holds. It records the reason for every conclusion, and stays unresolved where the evidence does not permit a claim.
- Synthetic BAM
- Truth-blind scan
- candidate observations
- Not truth classified
- Supplied synthetic event evidence
- Deterministic reasons
- Review status
- Independent verification
Discovery candidates joined to independent benchmark evaluation, built and run on development data, not on this page. Here the two checks run on separate fixtures and share no data: no discovery candidate appears in the Event Assurance check.
- Planted synthetic fixture
- Discovery inputs recorded from the run
- Deterministic rules
- Offline verification
Event Rescue: blind discovery
The scan reads a declared region of a synthetic BAM. It is never given the benchmark truth or the pipeline's VCFs, and it runs twice to prove its output is reproducible.
What the scan received
Read back from the recorded invocation and matched to discovery inputs by content.
- Generator
- Reference
- Bases scanned
Candidate observations
| ID | Locus | Change | Usable depth | Alt reads | Alt fraction | Fwd / rev |
|---|
Read support at this position
Evidence from the scan
- Scan inputs
- Generator
- Discovery manifest
- First run
- Repeat run
- Repeatable
Event Assurance
A separate check over a separate fixture. Comparator output here is a supplied synthetic fixture input: the benchmark comparator does not run. Read evidence is counted from the fixture's synthetic BAM. None of these events is a discovery candidate.
Each result compares the reads with the pipeline's own decision, so a filtered variant with strong read support is Contradicted too.
Pipeline disposition . Benchmark truth is not a WAVE policy input. Each reason is shown beside the counted value that made it hold. Not measured by this extractor: .
| Reason | Condition tested | Counted value |
|---|
Technical evidence is not benchmark truth
| Technical evidence assessment | ||
|---|---|---|
| Benchmark classification | ||
| TP or FP claim |
The policy assessed the counted technical evidence without the benchmark outcome. The benchmark layer then declines to classify an event outside its assessable region. A technical assessment, supporting or contradicting, permits a statement about the evidence only; it does not make the call a true or false positive.
Verification
An independent verifier recomputes the published package. A copy changed after certification is rejected, so review conclusions stay bound to the declared inputs and the published artifacts.
| Artifact | Result | Evidence |
|---|
- Profile
- Package
- Change made
- (
)
| Check | Original | Changed copy |
|---|
Claim boundary
Implemented here
Built outside this preview
Planned
Not claimed here
- Clinical accuracy
- Sensitivity or specificity
- Reduced review time
- Performance on laboratory or patient data
- Equivalence to a production variant caller