Public-data VUS-reclassification benchmark — 200 variants

Error-analysis writeup + replay improvements

Add reply draft for Jordan's kidney/HCT panel email

Lab-panel baselines: kidney 96.5%, HCT 85.4% + PKD2 BP4 cap

Fix author name: Theo Sevitt (not David)

Render lab brief as PDF + ship the converter

Add lab brief for Jordan Lerner-Ellis review

50-variant smoke test: RAG nets +7 / 0 regressions with tightened prompt

Smoke test: tightened RAG prompt fixes regression on 10 known cases

Partial RAG validation (596/1000) — net -11 vs deterministic baseline

Add per-gene / per-variant-type validation breakdown script

Add 1000-variant validation report (87.4% adjacent-tier, --skip-rag)

Add on-prem clinical deployment scaffold

Reach 89% deterministic concordance + add air-gap and governance scaffolding

Initial VariantLens clinical readiness scaffold