Research that defines the sovereign AI control plane.

Introduces PDDS as a research and engineering model for systems in which deterministic code, stochastic models, autonomous agents, policy engines, and human-machine actors coexist.

Introduces a protocol that turns proposed agentic mutations into governed intents evaluated against state, policy, execution bounds, and evidence requirements.

Separates stochastic reasoning from bounded execution authority in regulated and multi-organization environments.

Computes execution authority from structured, verifiable artifacts instead of standing credentials.

Treats a machine-enforceable protocol, rather than generated code, as the durable authority-bearing software artifact.

Introduces a certificate-bound runtime admission layer for autonomous execution authority.

Certifies the semantic safety of agent proposals using diverse, read-only validator panels and risk-adaptive quorum predicates.

Moves replication from bitwise state identity to evidence-backed belief lineage and bounded semantic coherence.

Provides a mandatory runtime enforcement point for certificate-bound authority at mutation time.

Replaces prose-first telemetry with signed, content-addressed state-delta evidence for autonomous operators.

When protocol-compliant agents agree, what prevents them from agreeing on the same wrong reality? Formalizes epistemic faults in protocol-compliant reasoning participants and separates semantic safety risk from liveness degradation.

How many genuinely independent epistemic paths support a quorum? Shows that nominally distinct agent reviewers can share upstream failure roots, so replica count alone does not establish epistemic redundancy.

What evidence and assurance must exist before a high-consequence action may execute? Makes action readiness an explicit, risk-conditioned evidence requirement separate from identity authorization.

When can a non-deterministic cognitive decision safely mutate durable system state? Coordinates data, evidence, policy, and delegated authority so committed effects remain compatible with their derivation inputs.

Documents a deterministic control-plane architecture that isolates unstable reasoning from live operations while preserving auditability.

Maps sovereign execution across national AI infrastructure through federated institutional policy and localized authority.

Specifies an independently enforceable boundary and programmable control plane that turns verified, action-specific assurance into narrow execution authority for consequential AI-agent operations.

How should reasoning resources be allocated when cost and latency are not the only objectives? Allocating model capability, context, reasoning depth, verification compute, evidence budget, latency, and cost to achieve a required level of inference quality.

Which information can be removed without violating future reasoning or system invariants? Formulates safe forgetting, memory compression, and retention value without violating future reasoning or system invariants.

Which observations, transformations, and prior beliefs actually caused a decision? Distributed tracing for beliefs, evidence, transformations, and decisions in non-deterministic systems.

Where should reasoning occur to balance independence, cost, latency, and authority? Placement theory for reasoning across fault, trust, provider, and sovereignty domains.

Which guarantees cannot simultaneously be maintained when agents must act despite incomplete or divergent evidence? Investigates fundamental impossibility tradeoffs between semantic consistency, autonomous availability, and operation under epistemic partition.

How should autonomous agents reach consequential decisions when they have distinct principals, incentives, budgets, utilities, and obligations?