Unisapience Labs produces original research, open benchmarks, and governance toolkits aimed at the gap between academic AI safety theory and enterprise deployment practice.
Timothy Poschel — Unisapience Labs
Filed as US Provisional Patent Applications 64/066,231 (SSPLX-001-PROV) and SSPLX-002-PROV. Discloses the n-simplex risk-decomposition framework, the shadow simplex construction, the seven-factor defeasibility-weighted SSS, and a structural extension comprising multiplicative product-form aggregation, modal register stratification, capability normalizer C(κ), compositional periodicity with explicit alignment-edge exclusion, and a transcendental meta-condition veto layer.
SSPLX-001-PROV
US Provisional 64/066,231
Filed: May 15, 2026
SSPLX-002-PROV
Companion application
Filing window: May 2026 – May 2027
Timothy Poschel — Unisapience Labs
A mathematical and conceptual tool for analyzing pathological attractors in self-evolving AI systems. Identifies five fundamental pathologies, maps their ten pairwise couplings and ten higher-order emergent dysfunctions onto a 4-simplex topology, and proposes 35 testable hypotheses with concrete experimental protocols. Foundation work for the patent-pending SSPLX scoring methodology above.
Figshare DOI 10.6084/m9.figshare.30223396
timposchel.com (HTML pre-print)
Published: December 1, 2025
Scalable oversight mechanisms, constitutional AI failure patterns, and red-teaming frameworks for enterprise-deployed systems. Current focus: the relationship between in-distribution reward maximization and out-of-distribution capability brittleness.
Multi-agent orchestration safety, tool-use evaluation, and autonomous workflow failure characterization. Enterprise focus: action cascade risk, agentic drift in production pipelines, and cross-system coherence under real deployment conditions.
Formal mathematical foundations for AI governance scoring — treating risk as a measurable, auditable property rather than a narrative judgment. Developing the SSS scoring system as a FICO-analog for enterprise AI.
We believe enterprise AI safety requires community-wide visibility into evaluation standards. These resources are released under open licenses.
An enterprise-oriented evaluation suite for AI safety properties. Tests cover agentic drift detection, cascade failure propagation, cross-system coherence measurement, and shadow AI surface area — specifically targeting production-scale deployments rather than research settings.
AI Bill of Materials schema for documenting third-party model components, training data sources, fine-tuning lineage, and tool integrations. Designed to interoperate with SBOM formats and EU AI Act Article 13 documentation requirements.
{
"aibom_version": "1.0",
"model_id": "acme-underwriting-v3",
"base_model": { "provider": "OpenAI", "version": "gpt-4o" },
"fine_tuning": { "dataset": "internal-claims-2023", "pii_review": true },
"tools": [{ "name": "claims-api", "version": "2.1.4" }],
"risk_tier": "II",
"sss_score": 74
}
Pre-built templates for AI governance documentation: model lifecycle policy, incident classification rubric, vendor AI risk assessment questionnaire, and EU AI Act Article 9 risk management system template.
Pre-trained classifiers for detecting specific SSS failure patterns in model outputs and system logs. Intended as lightweight monitoring primitives that teams can integrate into existing MLOps pipelines.
Running the 35 testable hypotheses from the pre-print against real-world enterprise AI system data. Seeking collaboration partners with access to production agentic systems.
Developing standardized normalization procedures so SSS scores are directly comparable across model families, sizes, and deployment contexts — a prerequisite for industry-wide adoption as a standard.
Extension of the Shadow Simplex to FDA-regulated AI/ML-SaMD contexts. In preparation — awaiting clinical validation data. The mathematical correlation between AI failure modes and clinical decision support system failure patterns shows structural parallels worth formalizing.
Development of the five-aspect decomposition method (Pentarchic Theory) as a general analytical framework for complex system failure analysis — the intellectual foundation underlying the 4-simplex structure of the Shadow Simplex.
We welcome collaboration with academic institutions, enterprise practitioners, and independent researchers working on AI safety, governance mathematics, and agentic system evaluation.
We're seeking organizations willing to run the UniSafety benchmark against their production AI systems and share (anonymized) results. All participating organizations receive a complimentary preliminary SSS assessment.
Researchers in RL theory, multi-agent systems, and AI governance mathematics are invited to contact us. We're open to co-authorship on empirical validation work.
We engage with regulatory working groups on AI governance standards. The SSS framework is designed to be adoptable as an industry standard — contact us about participation.
The Shadow Simplex framework isn't just a theoretical exercise — it's the backbone of every AI Risk DD engagement we deliver.