Every audit point is validated twice: once as deterministic fact in the MAE Forge (lab instruments, FAT-grade evidence), and once as governance fact in the clinical ward (MScAIMed sandbox, real care settings). The Sovereign Architect holds the blueprint, issues the OEM specification to the Forge, and receives the validated hardware at Handover as Clinical Validator.
OEM-funded fabrication and validation at NTU MAE facilities. Instrument-checkable evidence: oscilloscope traces, motion-capture metrology, ladder-logic verification, WORM commit records. Output: FAT-grade acceptance package.
MScAIMed clinical settings. The handed-over hardware is validated under real care workflows: patients, staff, ward acoustics, shift cycles, infection control. Output: clinical evidence file for the HSA dossier.
Governance-interposer validated on a lab arm: clamp latency and gate timing measured against blueprint bounds.
Gate behaviour verified under real care workflows: clamp events logged during assistive tasks, zero unsanctioned actuation across the observation period.
Bench validation of the sensing-to-alert pipeline: voxel chain latency and alert assertion measured end-to-end.
Confirms no identifiable imagery can exist in occupied wards (privacy audit of raw sensor output) and measures alert sensitivity/specificity with real patient movement and nurse response times.
Mobile-base surrogate validates constraint-bounded velocity and heartbeat-loss stillness against blueprint limits.
Navigation among real patients and ward traffic; stillness-on-heartbeat-loss demonstrated during live operations with zero contact incidents as the pass criterion.
Conformity to positional tolerance ±0.05 mm proven under motion-capture metrology.
Metrology re-verified in situ after transport and installation; tool-positioning conformity confirmed in clinical procedure conditions, sterile-field compatible.
Oscilloscope-verified bounds of the mandatory deliberation window. Officially noted as Audit Point 2 in the register.
Clinicians confirm the pause is perceptible, predictable, and non-obstructive within care routines; window integrity re-checked in situ under ward power and thermal conditions.
Bench proof of power severance <1.05 ms (design-intent), brake assertion via safety-stop channels, and the non-networked Fail-Loud alarm.
Nursing-staff stop drills in patient-adjacent (simulated) scenarios; Fail-Loud alarm audibility verified against real ward acoustics; brake assertion demonstrated under operational load.
Ladder-logic verification of three-factor human authorization; frames the Sovereign-In-The-Loop doctrine.
Authorization sequence executed by actual clinical staff: usability under gloves and time pressure, and observed lockout of every unauthorized action attempt.
Commit-to-finality integrity across system resets — unalterable forensic records proven on the bench.
Ledger continuity demonstrated across ward shift cycles, power events, and cleaning regimes; the WORM record accepted as the clinical incident log of record.
Macro-level safety proven by measured absence of surplus compute — the cycles required for model hallucination physically do not exist.
Sustained ward duty at peak load without cap breach; advisory outputs remain deterministic under the busiest clinical conditions — no drift, by physical impossibility.
OEM spec compliance and audit competence: manufacturing control validated at the source, per the standing OEM rule.
Modules integrated into the ward: mounting, infection-control compatibility, total IT-isolation verified in the hospital environment, and care-staff operational acceptance recorded.
Two evidentiary claims, independently witnessed, converging on one submission. Specification flows from the blueprint to the Forge; validated hardware flows back at Handover; the clinic completes the governance record. The blueprint never leaves the Architect's hands.
The governance mechanism works exactly as engineered — deterministic, instrument-checkable, FAT-grade. Signed by the Architect as issuer of specification; witnessed by OEM acceptance records.
The governed device is safe for patient use — proven in real clinical settings. Received and signed by the Clinical Validator; witnessed by clinical staff and ward records.