Verity classifies variants, resolves pharmacogenomic phenotypes, and computes doses with deterministic, guideline-cited engines, sealed with a recomputable hash and independently re-verifiable. A CAP inspector, a journal reviewer, and a skeptical MD can each recompute any output offline.
Product status and scientific standing are separate. Beta describes how finished the software is. It says nothing about the evidence behind the model, and the research-use-only boundary is not a beta limitation that general availability will lift. What has and has not been evaluated.
⚠ The categorical and points engines disagree here, so Verity flags this for human review instead of picking a tier.
Richards 2015 combining rules + Tavtigian 2020 points. The same functions run in the product, on the server, and in the known-answer test suite, with identical output everywhere.
ACMG/AMP combining rules (Richards 2015) plus the Tavtigian points model, transcribed from the papers and known-answer tested. The model orchestrates and explains; it never emits a code or a number.
Every applied criterion must carry an evidence source: a PMID, a ClinVar SCV, a gnomAD frequency. Enforced in the engine and again in the database. No un-sourced codes, ever.
Each classification is sealed with a SHA-256 hash over its evidence and verdict, on a hash-chained audit trail. Anyone can re-derive it, and the platform can refute a verdict that doesn't reproduce.
Richards categorical and Tavtigian points run in parallel. When they disagree, Verity flags it for molecular-pathologist review instead of quietly choosing a tier.
A classification is locked and requires molecular-pathologist sign-off. Research-use-only, non-diagnostic, physician-in-the-loop, designed to sit inside the FDA non-device CDS boundary.
ProtParam-class physicochemistry (pI, MW, ε₂₈₀, GRAVY, aliphatic index), a developability liability scan for biologics, and codon-optimized construct design: all deterministic, verified against published ExPASy values.
A Verity classification hands off to a real wet-lab assay in Olto (organoid or plate → dataset → back-calculated potency = functional evidence) and returns, closing interpretation → validation → evidence.