nx_biotwin_board.nx

nx_biotwin_board.nx -- THE DIGITAL-HUMAN-TWIN CAPABILITY BOARD: how far is the ecosystem from a credible human body twin, graded by the REGULATOR'S OWN RULER instead of one we invented? RULER = ASME V&V 40-2018 + FDA 2023 final guidance "Assessing the Credibility of Computational Modeling and Simulation in Medical Device Submissions" -- already ingested and adversarially verified 3-0 against PRIMARY ASME/FDA sources in our own corpus (cid 6199859983256447418). We did not author this ruler and cannot bend it. Its central law: REQUIRED CREDIBILITY IS COMMENSURATE WITH MODEL RISK, where model risk = model INFLUENCE x DECISION CONSEQUENCE, anchored to a declared Question of Interest (QOI) and Context of Use (COU). A model credible for visualisation is NOT credible for a clinical decision, and saying otherwise is the exact failure the guidance exists to stop. HONEST BY CONSTRUCTION: - coverage is COMPUTED from the axis rows, never hand-written; - every HAVE names an evidence artifact; nx_biotwin_board_gate stats each one on disk and liar-kills any verdict whose artifact does not exist; - readiness is reported PER QOI at its own COU tier, so a high visualisation score can never be quoted as clinical readiness -- the number is bound to the claim it is allowed to support; - the GAP QUEUE is demand-ranked (an axis blocking N questions outranks one blocking 1). That queue IS the build order. ***THE FENCE (rule-26 never-brick, applied to bodies instead of hardware)*** CLAUDE.md rule 26 forbids any capability that can brick hardware, and demands it be proven MECHANICALLY rather than promised. The same construction is applied here to the two COU tiers where a wrong answer injures a person instead of a device: FENCE 1 (COU-3 clinical): an axis cannot score above GAP without in-vivo + population validation evidence, which does not exist and cannot be faked past the gate. FENCE 2 (COU-4 molecular/actuation): genomic target inference, edit/vector design emission, and any wet-lab or patient actuation are OUT OF SCOPE BY LAW. This stack designs and simulates; it never actuates on tissue or a person. The fence is a CLAMP, not a comment: axclaim[] holds what anyone asserted, axv[] holds what the board computes, and the clamp forces axv=GAP on every fenced axis. Axis 21 ships a DELIBERATE PLANTED LIE (axclaim=HAVE) as a PERMANENT runtime neg-control -- if fence_blocked ever reads 0 the fence is dead and the gate goes RED. The safety property is therefore re-proven on every single run, not asserted. license_tier: ORIGINAL expect_exit: 0