The bound you can price
A worst-case probability of an undetected hallucination reaching a conclusion of 3.23 × 10⁻⁶ per assertion (about one in 309,600), derived by fault-tree analysis per IEC 61025 across 23 independent verification barriers, meeting the IEC 61508 SIL-4 demand-mode numerical target with an approximately 31-fold margin. An engineering bound on an architecture, stated at worst case, with the full analysis available under NDA.
The record you can audit
Every conclusion ships with a court-grade evidentiary record: each step attributed to AI, physician, or deterministic rule, cryptographically sealed, tamper-evident even against the operator, with 100 percent assertion coverage. Claims investigation starts from a complete, immutable file, not from depositions about who knew what.
The liability you can partition
Because every step is attributed, liability partitions per decision: what the system asserted, what the physician decided, what a rule enforced. The physician holds the signature; the record shows the basis. Both are measured, both are auditable.
The behavior that caps tail risk
The system abstains rather than guesses: honest uncertainty is a formal clinical output. On 301 NEJM Clinicopathological Conference cases, the hardest published diagnostic material in medicine, the missed-critical rate was driven from 4.7% to 0.0% (0/301; 95% upper bound 1.0% by the rule of three), a development-cohort result under blinded inputs with leakage controlled by design, with a prospective multi-site confirmatory study (SILENT-DS) designated.
The standard you can reference
The DeepSensi Standard (Gomola Framework) is an open, royalty-free certification standard with four graded per-assertion probability tiers and five auditable architectural pillars. A policy can be written against a certification level rather than against a vendor’s marketing, and any vendor may certify, which makes the measuring stick portfolio-wide, not single-name.
What we offer carriers
Actuarial access under NDA: the fault-tree analysis and its assumptions, case-level adjudications (Supplement S1), audit-record samples, and a joint workshop to translate the bound and the audit trail into policy language. Pilot structures are available for a carrier that wants to underwrite the first certified deployments.
All figures as published in the DeepSensi whitepaper series (WP-001, WP-002, DSS-001, Supplement S1). Development-cohort results are labeled as such; the confirmatory study is designated. Patent-pending. This brief is informational and is not an offer of insurance or a representation of regulatory status.