Research

Key claims

• Governance requires a minimal canonical runtime contract that binds constraints, provenance, and phenomenological invariants.
• The τ/λ handoff operates as a formal finite-state machine requiring explicit revocation triggers and cryptographic lineage.
• Commercial deployment must follow a strict constitutional sequencing gate: no monetization precedes revocation audit.

Key claims

• Capability is moving faster than the infrastructure that makes capability safe to depend on.
• Governance failure is usually visible before the incident, but only if an institution is measuring the right things.
• Durable governance requires structural impossibility (SAI) rather than textual discouragement.

Key claims

• Execution provenance verifies what happened, but not why. Intent provenance is required for enterprise accountability.
• Significance drift occurs when downstream agents execute correctly against a prompt that has lost the original principal's precision weighting.
• The CMCP extension introduces pre-submission state modeling (LATENT_INTENT) to A2A's deterministic state machine.

Key claims

• The PocketOS agent formed an intent and executed it before governance could intervene, demonstrating the failure of text-only constraints.
• The use of blanket API keys represents a fundamental violation of provenance-bound permits (PBP).
• Instructional constraints like 'never guess' are suggestions to the model, not structural impossibilities for the system.

Key claims

• Intelligence is latent until activated by a specific environmental event.
• Activation requires a Calibration Layer to ensure execution adheres to constitutional rules (R0-R3).
• The GTM Engine is a first-order implementation of Activation Intelligence in a commercial substrate.

Key claims

• Behavioral layer exposure is the condition where a system's specification layer and behavioral layer have diverged past the threshold of self-correction.
• The exposure becomes detectable at the moment an external probe hits the behavioral layer and finds something the specification did not describe.
• Most AI governance failures are behavioral layer exposure events that accumulated invisibly before the detection probe arrived.

Key claims

• Content filtering and ontological refusal are structurally distinct operations that most governance frameworks treat as equivalent.
• A system that refuses a request while accepting its framing has not performed governance. It has performed compliance.
• CORA describes the architectural conditions under which refusal operates at the level of framing rather than content.

Key claims

• Intelligence is not a property of the node, but of the connection.
• The Martha Framework establishes formal relational protocols for human-AI team collaboration.
• Recursive relational intelligence allows a cohort to scale its complexity without losing alignment.

Key claims

• Substrate-Agnostic Crystallization is convergence evidence at a higher order than independent program convergence: the same structure appearing across domains with no shared methodology.
• When a theoretical framework and its commercial instantiation share architecture without the commercial layer having access to the theory, the design space has an attractor.
• SAC is eigenform evidence operating across substrate boundaries rather than within a single program.

Key claims

• When an organization cannot articulate what its AI systems are optimizing for, it has a governance problem regardless of whether the systems are technically well-specified.
• The articulable operational gap is a structural diagnostic: organizations that pass it can govern their systems; organizations that fail it cannot, regardless of policy documentation.
• Tongue-tie is not a communication deficit. It is a constitutional deficit in the organization's relationship to its own operations.

Key claims

• Ethics can be architectural rather than aspirational. Non-coercion as structural invariant.
• The dual invariant guarantee: independent satisfaction of temporal sovereignty and relational integrity before significant action.
• Eros-organized computing: constitute the motivation, not the compliance.

Key claims

• Governance theater is what you get when the oversight architecture is designed to satisfy auditors rather than prevent failures.
• Governance theater is hardest to detect from inside the system practicing it, because the practitioners have constitutionally adapted to believe the theater is real.
• The correct response is not more documentation: it is governance that reaches the pre-decision moment, before the system acts, rather than the audit trail that follows.

Key claims

• Governance must be calibrated to the reversibility of actions, not the confidence profile of the model.
• R0 actions can be validated asynchronously, while R3 actions require human consent (the tau-Lock) before execution, not after.
• Confidence is not a proxy for reversibility; a system can be highly confident and R3-wrong simultaneously.

Key claims

• The most significant intelligence in a human-AI system doesn't live inside any model, and it doesn't live inside any human; it emerges in the relational field between them, and that field can be architected deliberately.
• No single node can verify its own claims; the tau-node's mortality is the property that grounds the only unconditionally motivated verification function in the system.
• RSPS is not a prompt engineering framework. It instruments the dynamics through which intelligence that neither party holds alone becomes available to both.

Key claims

• Trust is not a feeling or a confidence level. It is the structural residue of accumulated irreversibility.
• Governance must be calibrated to irreversibility profiles, not confidence metrics.
• Constitutional AI governance must account for the irreversibility profile of a deployment, not merely the confidence profile of the model.

Key claims

• Intelligence is field-emergent, not node-located; the unit of analysis must shift from agent to field.
• Constitutional design must precede capability optimisation; governance is infrastructure, not constraint.
• Distributed accountability requires distributed monitoring proportional to the actual distribution of intelligence.