OncoRef 1.26.0: reviewed normal-cell evidence and Other audit

OncoRef 1.26.0 adds reviewed rare-normal-cell and activated-immune evidence to the independent normal-safety assessment, with a complete audit of all 600 previous Other proteins.

Category	Proteins
CTA	701
Unsafe	18,062
Other	454
Unassessed	646

All 698 previous CTAs remain. CDY1/B, OR2T7 and BPY2/B/C are added. Of the 600 baseline Other proteins: 3 become CTA, 31 Unsafe, 112 receive review holds and 454 remain Other.

Normal safety, cancer presence and reproductive identity remain independent. Existing RNA/IHC safety ladders are unchanged. Any positive measured tumor RNA now attests to cancer presence; reliable reproductive protein supplies presence independently of the RNA-only floor. Neither overrides safety, concentration or required coverage.

gene_knowledge.findings() exposes 210 accepted, source-linked findings with exact identities, tissue/cell/state, native units and limitations. All 611 baseline Other genes were searched; search retrieval is distinct from individual paper adjudication. The human olfactory supplement uses the shared pinned reference/cache API.

cta_selection.sequence_dispositions() provides direct-target and shared-sequence screening roles: CTA candidate; Unsafe exclude; Other allow under the named policy; Unassessed review. These do not establish peptide/HLA clearance or update the independently versioned legacy FASTA. MROH5/SERPINA2 translation QC remains tracked in #657; mapped identity is not a complete-translation guarantee. Human rules do not transfer to dog/mouse by orthology.

Release guide · Complete 600-protein audit · Funnel and logic · Rules/API · Reviewed PR #656

Attached: the 31-page protein audit, 36-slide logic PDF and editable PPTX, 9-page TCR-T safety report, per-protein decisions, reviewed findings, native normal measurements, 143-cohort cancer audit and package artifacts.

Complete and cached full-context replays are byte-identical. Artifact hashes, all 600 decisions, high-raw holds, package API, documentation and rendered reports were verified. Validation: 2,836 local tests passed / 22 skipped; all eight CI checks passed, including Python 3.10–3.14 and Windows. Detailed results are recorded in the PR.