Mechanism

The affective state field is the seventh structural field of the semantic agent schema, alongside the intent field, context block, memory field, policy reference field, mutation descriptor field, and lineage field. It is a deterministic, policy-bounded data structure that encodes valence-weighted feedback derived from prior execution outcomes and environmental observations. It does not encode emotion in the phenomenological sense. It encodes a structured modulation vector that influences how the agent weighs alternatives, tolerates ambiguity, persists under partial failure, and escalates under constraint pressure.

What makes the encoding deterministic is the update function. Given the same agent state, the same environmental inputs, and the same policy configuration, the affective state update function produces the same output. The update function is implemented as a deterministic state transition function: it takes the agent's current affective state vector, the current set of structured observations, and the applicable policy configuration, and produces an updated affective state vector. Each dimension of the vector is updated independently according to its own update rule, subject to policy-imposed bounds. The update is recorded in the agent's lineage as a state mutation event, with the input observations and the resulting state change preserved for audit.

Determinism is the property that ties affect into the same governance, lineage, and policy machinery that applies to every other agent field. Because the affective state field is a structural field and not metadata, an annotation, or an external signal, every mutation to it is recorded in lineage, subject to policy validation, and auditable by governance infrastructure. This is what distinguishes the disclosure from systems that treat affective state as a side-channel, a prompt modifier, or an external behavioral overlay.

The Structured Modulation Vector

The thing being encoded deterministically is a structured modulation layer comprising a plurality of named control fields, each a distinct dimension of the agent's dispositional orientation with defined semantics, value ranges, update rules, and governance bounds. The disclosed named control fields include an uncertainty sensitivity field, an ambiguity tolerance field, a novelty appetite field, a persistence-under-partial-failure field, an escalation-under-time-pressure field, a risk sensitivity field, and a cooperation disposition field. These are not arbitrary labels; each corresponds to a measurable modulation axis.

Each named control field is represented as a tuple comprising a current magnitude value within a defined range, a decay rate governing how rapidly the field returns toward a baseline in the absence of reinforcing stimuli, a policy-defined ceiling and floor bounding the permissible range, and a timestamp recording the most recent update. Because each dimension carries its own bounds and decay parameters and is updated independently, the vector is independently readable, writable, and auditable field by field, and the same observation can move different dimensions at different rates without coupling them through a shared scalar.

Observations That Drive Updates

The update function operates on structured observations derived from the agent's execution environment, not on raw signals. The disclosed observation types include repeated failure patterns, for example a sequence of execution outcomes matching a failure signature such as consecutive mutation rejections at the same validation gate; competing objectives, where the intent field or planning graph holds multiple active objectives with conflicting resource requirements; time pressure signaled by an approaching deadline or a narrowing execution window; novelty exposure, where inputs fall outside the agent's historical distribution as determined against the memory field; uncertainty levels reported by an inference engine as low confidence or high entropy; and execution success patterns, where completed steps move the agent toward a more confident disposition.

Each observation type is admitted only for the fields it is permitted to drive. The mapping from observation to dimension is part of what the deterministic update consumes: a repeated-failure observation increases uncertainty sensitivity and risk sensitivity and adjusts persistence depending on whether the failures are of the same type or varying types; a novelty-exposure observation elevates or suppresses novelty appetite depending on whether prior novelty exposures correlated with positive outcomes or failure. Because the observation set is structured and recorded, the same inputs replayed under the same policy reproduce the same trajectory.

Policy-Bounded Update Mechanics

Every update is a policy-bounded mutation. The policy reference field specifies, for each named control field, a set of constraints: range bounds giving a minimum and maximum permissible value; rate limits giving a maximum magnitude of change per update cycle; admissible triggers giving the observation types permitted to drive that field; update authority giving which entities or processes may initiate updates; and decay governance constraining the decay parameters.

These constraints execute as a defined sequence. When a structured observation is received, the update function first verifies that the observation type is in the admissible trigger set for the relevant field; then computes the raw update magnitude according to the field's update rule; then clamps the update magnitude to the rate limit; then applies the clamped update to the current field value; then clamps the resulting value to the range bounds; and finally records the complete transaction, the observation, the raw update, the clamped update, the prior value, and the resulting value, in the agent's lineage. This multi-stage clamping ensures that no single observation, and no sequence of observations, can drive the affective state outside its policy-defined operating envelope. If the computed update would exceed the ceiling, it is clamped to the ceiling; if it would fall below the floor, it is clamped to the floor.

Decay, Hysteresis, and Stabilization

Each named control field is governed by an emotional decay curve that determines how the field returns toward its baseline in the absence of reinforcing stimuli. The decay curve is a deterministic function of the time elapsed since the most recent update, the magnitude of the current deviation from baseline, and the decay parameters from policy. In an embodiment the decay function is an exponential decay with a configurable time constant, V(t) = V_baseline + (V_current minus V_baseline) times exp(minus t over tau), where tau is the policy-specified time constant. Different fields may carry different time constants: uncertainty sensitivity may decay rapidly because epistemic conditions change frequently, while persistence-under-partial-failure may decay slowly because learned persistence reflects deeper accumulated experience.

The layer exhibits semantic hysteresis: the current affective state depends not only on the current observations but on the trajectory of prior states. Hysteresis is implemented through asymmetric update rules, where the rate at which a field increases in response to a triggering observation may differ from the rate at which it decreases when the triggering condition is removed. In an embodiment, negative-valence updates driven by failure, uncertainty, or threat apply at a higher rate than positive-valence updates driven by success or stability, producing a built-in caution bias.

Entropy-governed valence stabilization prevents oscillatory behavior. When a field exhibits rapid alternation between elevated and suppressed values, detected by monitoring the frequency and direction of recent updates, the stabilization mechanism progressively increases the effective decay time constant, damping the oscillation. The stabilization threshold and damping factor are policy-configurable. None of these stages introduce nondeterminism: decay, hysteresis, and stabilization are all deterministic functions of recorded inputs and policy parameters.

Encoding Does Not Override Governance

A strict separation of concerns is maintained between the affective modulation layer and the governance infrastructure. The affective state field cannot create authority the agent does not possess, cannot bypass policy constraints, cannot validate truth claims, and cannot authorize execution that governance has denied. It modulates how the agent thinks, not whether the agent is permitted to act. The separation is enforced structurally: the affective state field is not an input to the governance gate, which evaluates execution admissibility from policy compliance, trust slope validation, and cryptographic provenance independently of affective state.

This matters for the determinism property. Because affect is encoded as bounded, recorded, replayable state rather than as a free-floating influence on admissibility, even a maximally confident, minimally risk-sensitive affective configuration cannot relax a policy ceiling, grant a permission, or treat uncertain information as verified. The deterministic encoding gives affect a precise, auditable place in the pipeline without letting it become a vector for circumventing governance.

Forensic Reconstruction From Lineage

The payoff of deterministic encoding is forensic reconstruction. Because every update is deterministic and every driving observation is recorded, the agent's affective state at any historical point is reconstructable by replaying the update function over the sequence of recorded observations from lineage. The reconstruction produces the exact affective state vector that existed at the queried timestamp. This enables compliance auditing and regulatory review of an agent's disposition at the time of a disputed decision without requiring persistent storage of moment-to-moment affective values: the reconstruction is performed on demand from the lineage record and the update function specification, both preserved as part of the agent's cryptographic provenance.

The lineage records reference affective mutations abstractly, recording the observation type, the update direction on each affected field, and the policy compliance status, rather than absolute field values or raw observations. This permits verifying that the agent's affective evolution followed policy-compliant paths while preserving the privacy of the agent's moment-to-moment state, since the absolute values can be re-derived deterministically when an authorized audit requires them.

Disclosure Scope

The deterministic encoding of affective state, comprising the structured modulation vector of named control fields each represented as a magnitude, decay rate, policy-defined ceiling and floor, and update timestamp; the deterministic state transition function that maps the current vector, the structured observations, and the policy configuration to an updated vector; the multi-stage policy-bounded update sequence of admissible-trigger verification, raw computation, rate-limit clamping, application, range-bound clamping, and lineage recording; the deterministic decay curves, asymmetric hysteresis, and entropy-governed stabilization; the structural separation from the governance gate; and the forensic reconstruction of historical affective state by replaying recorded observations through the update function, is disclosed in the cognition filing (U.S. Application No. 19/647,395 and its international counterpart). This article describes that disclosed mechanism. The disclosure does not claim particular named control fields, particular decay constants, or particular policy contents; those are configuration choices made by adopters within the disclosed framework. An alternative embodiment permitting a bounded, policy-bounded, lineage-audited stochastic component is within scope, provided it does not compromise the governance properties described.