Primary technical disclosure

Permitted Deviation: Bounded, Recorded Departure From an Agent’s Declared Values Permitted deviation lets a persistent semantic agent depart from its signed policy object under a computed deviation likelihood, recording the departure as sanctioned rather than as fault, with a declared deviation deductible and aggregate retention bounding accumulated undischarged harm.

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.

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.

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..