Publications
Read the mathematics yourself.
Everything DeepSensi claims in public is written down, referenced, and offered for attack. Preprints are headed to medRxiv; manuscripts are in submission to peer-reviewed venues.
WP-001 · Reliability engineering
Formal Reliability Analysis of Multi-Layer Deterministic Verification in Clinical AI: A Fault Tree Approach
The safety case: 23 independent barriers modeled per IEC 61025, common-cause adjustment, worst-case bound 3.23 × 10⁻⁶ per assertion, corrected demand-mode SIL analysis, empirical NEJM CPC validation (N=301: 86.0% top-1, 93.7% top-3, 0.0% missed-critical), degraded-mode analysis, and the Gomola Framework certification levels.
Read · print to PDF →
WP-002 · Evaluation science
The Flawed Yardstick: Why Static Medical Benchmarks Penalize Clinical Safety
The Safe Triage Paradox, documented defects in HealthBench Hard (case IDs 0b8f1d60, 9ab66439), the EBM Safe Override, and Auto-CSA — a dynamic, safety-first evaluation protocol validated on N=301 NEJM CPC cases.
Read · print to PDF →
WP-003 · Diagnostic architecture
The Hyper Consilium: Integrating Human Clinicians as Scored Nodes in a Fractal AI Diagnostic Swarm
The diagnostic core: 37 specialist agents in adversarial triads, dual-pathway reasoning, a unified reputation economy that scores physicians as cognitive partners, outcome-gated physician royalties, and the LIMBO structured-uncertainty protocol.
Read · print to PDF →
WP-004 · Reputation science
The Global Clinical Performance Score: Objective, Bias-Resistant Physician Reputation
A five-axis, temporally decaying, anti-gaming framework for measuring clinical cognition — the first global meritocracy of medical talent.
Read · print to PDF →
WP-005 · Autonomous research
Autonomous Bayesian Hypothesis Generation and Adversarial Validation: The AutoResearcher-HYPO Architecture
An autonomous research engine with a formal Bayesian convergence guarantee, an adversarial validation swarm, and deterministic biological-network anchoring — so no model can inflate a result with fabricated mechanism. Running in ongoing public-benefit research missions; no biomedical findings are claimed.
Read · print to PDF →
WP-006 · Compassionate access
Golden Horizon: AI-Mediated Compassionate Access Through an Autonomous Agent Architecture
Borderless clinical-trial matching and autonomous patient advocacy — free for patients by architectural mandate, with anti-hallucination-verified matching, a cryptographic consent escrow, and a track structurally independent from commercial recruitment.
Read · print to PDF →
WP-007 · Digital epidemiology
GERN: Privacy-Preserving Syndromic Surveillance from Ambient, Verification-Grade Clinical Documentation
An always-on outbreak radar produced as a by-product of verified documentation: diagnostic-grade syndromic signal, irreversible de-identification, k-anonymity (k ≥ 5) and a differential-privacy budget, federation without raw-record movement, and public-domain detection statistics — aligned to the IHR (2005). A methods paper with a designated, pre-registered prospective evaluation; no detection results are claimed.
Read · print to PDF →
DSS-001 · Standard definition
The DeepSensi Standard: A Quantitative Safety Certification Standard for Clinical AI Systems
Five auditable pillars, four certification levels with explicit hallucination bounds, two-dimensional certification (architecture · probability), assessment methodology, EU AI Act mapping, and governance. Open and royalty-free.
Read · print to PDF →
Perspective
Why Medical AI Needs a Quantitative Safety Certification
The argument in three pages: every safety-critical industry measures reliability before deployment — clinical AI must too.
Read · print to PDF →
Auditors, editors, and qualified press may request the NEJM CPC case-level protocol and supplementary materials: [email protected]. Version 3.0 (July 2026) is the canonical edition of every paper. The methodology is protected by 12 filed patent applications, with a further 24 prepared — the majority designated for open publication as technical whitepapers.