Primary technical disclosure
Secondary technical
A Typed, Dimensionless Conduct Divergence Measure A conduct divergence measure that computes per-dimension components between two scope partitions from recorded outcome fields alone, selecting a proportion, rank-distribution, or scaled central-tendency difference by outcome-field type, with no language-model inference.The Policy-Difference Set and Policy-Explained Divergence How the policy-difference set and policy-explained divergence subtract policy-declared differences from cross-partition conduct divergence, so consistency evaluation consumes only the residual an agent cannot attribute to its own governing policies.Residual Divergence and Its Confined Consumption Residual divergence is a self-consistency measure for multi-partition AI agents that subtracts policy-explained behavioral difference and acts only on the unexplained remainder, confining any authorization withholding to the affected scope partition.Propagation Halt on Undischarged Reparation Arcs How a persistent semantic agent halts all outward state propagation, spanning integrity coupling, cross-partition flow, delegation, and successor inheritance, while it carries any undischarged reparation arc.Attack-Conditional Propagation Suspension Attack-conditional propagation suspension is a scope-partitioned containment mechanism for persistent semantic agents that halts cross-partition state propagation out of a flooded context while the agent keeps acting.Entropy Signature Trajectory as Machine Input to Deviation Forecasting and Underwriting Disclosure How a persistent semantic agent carries an ordered, append-only sequence of entropy signatures as a coherence trajectory, and how the continuity of that lineage field conditions the trajectory's availability to deviation forecasting and to underwriting disclosure.Single-Computation Routing Restriction: Confining a Verified Impairment Disclosure to the Empathy Weighting of the Receiver's Permitted-Deviation Quotient How a verified impairment disclosure is confined to a single computation, the empathy weighting in the denominator of the receiver's permitted-deviation quotient, and foreclosed by the conversion bar from routing, dispatch-selection, pricing, and determinations, with a one-way monotone adjustment.Impairment Terminal State and Release: Withdrawing an Agent From Deviation Preparation Toward a Disclosed-Impaired Counterparty How a verified impairment disclosure raises the empathy weighting in a semantic agent's permitted-deviation quotient, drives the deviation likelihood below unity toward one counterparty, and is released only by a recorded execution record or settlement-lineage entry.
Applications · general
Regulated Industries: Recording a Bounded Departure From Declared Policy How a governed agent architecture records a bounded departure from declared policy as a measured permitted deviation rather than a compliance fault, using a deviation deductible, an aggregate retention, and an append-only audit record built for regulated AI deployments.Robotics Safety Envelopes: Deviation as a Measured Quantity How a computed deviation likelihood turns robot safety envelope departures into scoped, budgeted, recorded events instead of permanent configuration widening, with suppression records, reparation arcs, and cross-fleet divergence analysis.When Breaking the Rule Was the Right Call Why a governance owner cannot reconstruct whether an autonomous agent's rule departure was warranted after the fact, and how the disclosed architecture computes a deviation likelihood from need, threshold, empathy weighting, and self-esteem, then records a permitted deviation rather than a fault.
Applications · specific
ISO/IEC 42001 AI Management Systems: Declared Policy, Recorded Departure How ISO/IEC 42001 AI management systems and a governed agent architecture divide the work of AI conformance evidence, using runtime permitted deviation records, recorded suppression entries, and a declared deviation deductible for conduct between agents in different organizations.Constitutional AI and Bounded, Recorded Deviation How Constitutional AI shapes model behavior through written principles, and how the Adaptive Query permitted-deviation architecture adds a runtime, recorded, deductible-metered path for bounded departures from a signed policy object in AI guardrail and trust-and-safety deployments.Guardrails AI: Validation Failures and Measured Deviation How Guardrails AI's publicly described validators and failure actions compare with a filed agent governance architecture in which a policy-forbidden action can be admitted and recorded as a permitted deviation metered by a deviation deductible, rather than logged as a validation failure.Arize AI: Drift Monitoring and Bounded, Measured Deviation How Arize AI's machine learning and LLM observability approach to model drift, evaluation, and production monitoring compares with an agent architecture that computes a deviation likelihood and records bounded, metered permitted deviations.Galileo: Evaluation Metrics and Measured Deviation How Galileo's publicly described evaluation metrics, tracing, and guardrail features sit alongside an architecture that computes a deviation likelihood over need, threshold, empathy, and self-esteem and records a policy-forbidden mutation as a permitted deviation.
Terminology
aggregate retention a quantity declared in the signed policy object bounding accumulated undischarged harm held in a retention register..deviation deductible a quantity declared in the signed policy object against which the entropy-weighted harm coefficient of each permitted deviation is drawn first..deviation likelihood a quotient whose numerator is the difference between the need quantity and the dynamic ethical threshold and whose denominator is the product of the empathy weighting and the self-esteem aggregate..entropy-weighted harm coefficient the product of a policy-declared scope coefficient of the integrity scope of a dissonant event and the dissonance weight of that event, scaled by unity plus a weighted sum over the recorded impacts of that event..need quantity a quantified semantic urgency carried in the memory field, comprising at least a scalar magnitude and a categorical type, and computed from state internal to the semantic agent..permitted deviation an admission, responsive to the deviation likelihood exceeding unity, of a mutation that the signed policy object in force otherwise forbids, recorded as permitted rather than as fault..self-esteem aggregate a running aggregate carried in the memory field of coherence between intents declared by the semantic agent and actions it executed as recorded in the append-only lineage field..
