Variant evidence review
Interpret PP3 and BP4 as calibrated evidence, not predictor votes
Computational predictions can contribute evidence for or against a deleterious effect. Concordance among correlated tools is not automatically independent evidence, and a raw score should not be converted into a criterion from model memory.
Agent question
Which computational evidence and applied PP3 or BP4 strength are returned for this public variant?
Report what the contract returns
Preserve each returned predictor observation, calibration context where provided, applied criterion strength and provenance. If PP3 or BP4 is not applied, do not infer it from a score or a majority vote.
Avoid double counting
Multiple predictors may use overlapping training data or features. Reviewers should use calibrated, criterion-specific guidance and consider splicing evidence separately when relevant.
- Returned score and direction
- Applied criterion and strength
- Calibration or threshold provenance when publicly provided
- Potential dependence among predictors
- Variant consequence and splicing context
- Gene-specific specifications
Public references
Clinical and data boundary
Folklore Clinical Variant Interpretation MCP is published by Helena Bioinformatics. It accepts public variant-level queries only, not patient, phenotype, family, segregation or private case data. Its results are automated decision support for qualified professional review, not a diagnosis or treatment recommendation.