Study 45: which of nine billion answers a laboratory can act on A safety review of AlphaGenome Atlas, measured live on 1,200 real variants at two genes. The headline AVI score separates every one of 600 variants at both loci. The dense per-track scorers do not: splice-site usage returns 950 of every 1,000 values shared with another variant at HBB and 998 at CFTR, and 145 variants at HBB and 574 at CFTR carry an identical pair of splice-site values. All seven _ACTIVE scorers share more than their base scorer at both loci, which the wire cannot distinguish from coarseness by design. The shared values of the base scorers concentrate in the near-zero band a laboratory uses to clear a variant. Arm A needs no key: the genetic code settles 184 of 576 substitutions exactly, and over the assembly 27,917,828 variants have a protein consequence that is already certain, across a coding fraction of 101,167,089 variants — about one in ninety. Every figure re-derived from fresh downloads and fresh pulls; all four seals reproduce. No prediction value is read or published, no variant is named, no clinical claim is made. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>