The Affective State Field as a Seventh Schema Field
The semantic agent schema described in the cognition filing comprises six foundational fields: an intent field, a context block, a memory field, a policy reference field, a mutation descriptor field, and a lineage field. These six fields let an agent carry its purpose, situational awareness, accumulated history, governance constraints, evolutionary record, and cryptographic provenance as a single persistent object. They do not, however, encode the agent's dispositional orientation toward its own execution environment: the modulation state that shapes how the agent evaluates candidates, prioritizes mutations, or adjusts its responsiveness based on the cumulative outcomes of prior operations.
The disclosure introduces a seventh field, the affective state field, as a structural addition to the schema. 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 or subjective sense. It encodes a structured modulation vector that influences the agent's deliberation dynamics: how the agent weighs alternatives, tolerates ambiguity, persists under partial failure, and escalates under constraint pressure. The field is instantiated as one or more of a scalar valence value, a multi-dimensional vector encoding distinct modulation axes, or a structured record of named modulation fields each with magnitude, decay parameters, and policy bounds.
Introducing affective modulation as a structural field, rather than as metadata, an annotation, a prompt modifier, or an external behavioral overlay, ensures that it participates in the same governance, lineage tracking, and policy enforcement mechanisms that apply to every other agent field. Every mutation to the affective state field is recorded in the agent's lineage, subject to policy validation, and auditable by governance infrastructure. The field is persisted with the agent across execution cycles, delegation events, and substrate migrations, so the agent's modulation state is not lost when it moves between execution environments or is serialized for transport.
The Structured Modulation Layer
In an embodiment, the affective state field is organized as a structured modulation layer comprising a plurality of named control fields. Each named control field encodes a distinct dimension of the agent's dispositional orientation and governs a specific aspect of its deliberation behavior. The fields are not arbitrary labels; each corresponds to a measurable modulation axis with defined semantics, value ranges, update rules, and governance bounds.
The named control fields comprise at least: an uncertainty sensitivity field, encoding responsiveness to epistemic uncertainty, where elevation produces more conservative candidate selection and a greater tendency to defer or escalate; an ambiguity tolerance field, encoding the capacity to hold multiple plausible interpretations in parallel rather than forcing premature resolution; a novelty appetite field, encoding disposition toward engaging previously unobserved patterns, entities, or execution paths; a persistence-under-partial-failure field, encoding the tendency to retry, adapt, or reformulate rather than abandon a line of execution showing degraded confidence; an escalation-under-time-pressure field, encoding the tendency to escalate decision authority or request external input under temporal constraint; a risk sensitivity field, encoding the weighting of potential negative outcomes relative to positive ones; and a cooperation disposition field, encoding the tendency to favor collaborative strategies such as delegation, resource sharing, and multi-agent coordination over independent execution.
In an embodiment, 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 value returns toward a baseline absent reinforcing stimuli, a policy-defined ceiling and floor bounding the permissible range, and a timestamp recording the most recent update. Each field is independently readable, writable, and auditable, occupying a defined position in a fixed-schema structure.
What Affective State Modulates
The affective state field modulates specific, enumerated targets within the agent's deliberation and execution pipeline. The targets are not open-ended; each is a defined computational parameter whose value the field adjusts within governance bounds. The field does not create new capabilities, authorize new actions, or bypass policy constraints. It modulates the quantitative parameters that shape how existing authorized processes execute.
The modulation targets comprise at least: promotion thresholds, the minimum score or confidence required for a candidate mutation, execution path, or planning graph branch to advance to the next evaluation stage; search breadth, the number of candidate alternatives explored at each decision point; branch growth rates, the rate at which new speculative branches are generated during forecasting; decay rates for unpromoted candidates, the rate at which candidates not yet promoted are discarded from working memory or planning graph structures; escalation thresholds, the conditions under which the agent transitions from independent operation to delegation, escalation, or help-seeking; persistence parameters, the number of retry attempts, reformulation cycles, or alternative strategy explorations undertaken before declaring failure; delegation routing preferences, the agent's preference ordering among available delegates; and mutation acceptance thresholds, the stringency of the validation gate applied to mutations proposed by an external inference engine.
Each named control field maps to these targets in declared directions. Elevated risk sensitivity or uncertainty sensitivity raises promotion thresholds and mutation acceptance thresholds, requiring stronger evidence before candidates advance. Elevated novelty appetite or ambiguity tolerance increases search breadth and branch growth, causing the agent to consider more alternatives before committing. Elevated persistence-under-partial-failure slows candidate decay and increases retry counts. The escalation-under-time-pressure field directly modulates escalation thresholds, while the cooperation disposition field modulates how aggressively the agent seeks delegation partners and how it ranks them.
Deterministic Encoding and Update Mechanics
Updates to the affective state field are deterministic: given the same agent state, the same environmental inputs, and the same policy configuration, the affective state update function produces the same output. This ensures the agent's affective evolution is fully auditable, reproducible, and governable. In alternative embodiments the update function may incorporate bounded stochastic components, for example noise injection to simulate biological variance, provided the stochastic contribution is policy-bounded, auditable through lineage recording, and does not compromise the governance properties.
The update function operates on structured observations derived from the agent's execution environment. These observations comprise at least: repeated failure patterns, where a detected failure signature such as consecutive mutation rejections at the same validation gate increases uncertainty sensitivity and risk sensitivity and modulates persistence based on whether the failures share a type; competing objectives, where conflicting resource requirements or contradictory success criteria elevate ambiguity tolerance and escalation sensitivity; time pressure, where an approaching deadline or narrowing execution window raises escalation-under-time-pressure and may reduce search breadth; novelty exposure, where inputs falling outside the agent's historical distribution elevate or suppress novelty appetite according to whether prior novelty exposures yielded positive outcomes; uncertainty levels from model confidence, where low-confidence or high-entropy inference output elevates uncertainty sensitivity; and execution success patterns, where successful steps modulate the state toward increased confidence disposition scaled by the significance of the outcome.
The update function is implemented as a deterministic state transition function taking the current affective state vector, the current structured observations, and the applicable policy configuration as input and producing an updated vector. Each dimension is updated independently according to its own 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 resulting state change preserved for audit.
Affect 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, bypass policy constraints, override trust slope validation, validate truth claims, or authorize execution that governance has denied. It modulates how the agent thinks, not whether the agent is permitted to act.
This separation is enforced structurally. The governance gate evaluates execution admissibility based on policy compliance, trust slope validation, and cryptographic provenance independently of the agent's affective state; the affective state field is not an input to the governance gate. Even if affective modulation produces maximal confidence disposition and minimal risk sensitivity, the governance gate independently determines whether a proposed action satisfies all policy requirements.
The separation operates across specific dimensions. On authority, affective state cannot grant permissions: an agent with elevated cooperation disposition still cannot delegate outside its policy-defined delegation scope. On truth validation, affective state cannot validate or invalidate factual claims; the agent's epistemic state is maintained independently of its dispositional state. On policy compliance, affective state cannot relax policy bounds: policy-imposed ceilings, scope limitations, and governance requirements remain inviolable, enforced as hard constraints during every update cycle. On trust slope validation, the agent's cryptographic lineage trajectory is computed and validated independently of any affective configuration.
Policy-Bounded Updates
Every update to the affective state field is a policy-bounded mutation. The agent's policy reference field specifies, for each named control field, a set of constraints governing how the field may be updated. These comprise range bounds, a minimum and maximum permissible value to which any computed update is clamped; rate limits, a maximum magnitude of change per update cycle that prevents discontinuous affective jumps even under extreme observations; admissible triggers, a defined set of observation types permitted to drive updates to each field, so that observations outside the set are ignored and spurious or adversarial signals cannot modulate fields they should not affect; update authority, a specification of which entities or processes may initiate updates, restricting writes to the agent's own execution environment, governance-authorized feedback channels, and policy-defined delegation parents; and decay governance, constraints on the decay parameters that prevent adversarial suppression of affective response through artificially accelerated decay.
The policy-bounded update operates in stages. When a structured observation is received, the function first verifies the observation type is in the admissible trigger set for the relevant field, then computes the raw update magnitude per the field's rule, then clamps the magnitude to the rate limit, then applies the clamped update to the current value, then clamps the result to the range bounds, and finally records the complete transaction (observation, raw update, clamped update, prior value, resulting value) in the agent's lineage. This multi-stage clamping ensures no single observation or sequence of observations can drive the affective state outside its policy-defined operating envelope.
Each named control field is also governed by a decay curve determining how its value returns toward baseline absent reinforcing stimuli, as a deterministic function of elapsed time, current deviation, and policy-specified decay parameters. Different fields may carry different decay behavior, reflecting that some modulation dimensions are more persistent than others. Asymmetric update rules produce a built-in caution bias in which the agent responds more rapidly to deteriorating conditions than it recovers from them, and a stabilization mechanism damps oscillation when a field exhibits rapid alternation between elevated and suppressed values.
Cross-Primitive Integration: Confidence and Forecasting
The affective state field serves as a cross-primitive input to the confidence computation and the forecasting operations described elsewhere in the cognition filing, creating a feedback loop in which the agent's cumulative execution experience modulates its willingness to execute and its speculative planning behavior.
With respect to confidence, the affective state modulates the rate at which confidence decays and recovers. When uncertainty sensitivity and risk sensitivity are elevated, confidence decays faster, so the agent transitions sooner from executing mode to a non-executing cognitive mode; when suppressed, confidence decays more slowly, permitting the agent to sustain execution through periods of moderate uncertainty. Elevated persistence-under-partial-failure increases the confidence recovery rate. These dynamics are bounded by policy to prevent both runaway confidence and confidence collapse, so that an agent accumulating negative experience becomes progressively more cautious while one accumulating positive experience becomes progressively more willing to execute, each within governed limits.
With respect to forecasting, the forecasting engine reads the agent's current affective state when initializing a planning graph generation cycle. Novelty appetite modulates the branching factor, the number of speculative alternatives explored at each forecasting step; risk sensitivity modulates the pruning criteria, how aggressively low-probability or high-risk branches are discarded; and persistence-under-partial-failure modulates the depth, how many steps into the future the engine projects before terminating exploration. The agent's speculative planning behavior thereby reflects its accumulated experience, producing more conservative forecasts when experience warrants caution and more expansive forecasts when it warrants exploration.
Embodiments and Multi-Agent Operation
The affective state mechanisms are substrate-agnostic in logical specification but substrate-aware in implementation. In a centralized deployment the update function executes within the substrate's agent runtime with low-latency access to the agent's complete state. In a federated deployment each node executes updates locally, with cross-node propagation mediated by delegation events and governance-authorized broadcasts under additional attenuation. In a fully decentralized deployment updates execute locally on each device using only locally available observations, with an increased contagion damping factor to account for the lack of centralized spiral detection. In an embodied deployment the field receives structured observations derived from physical sensor data, where for example constrained operating conditions may elevate risk sensitivity.
When a parent agent delegates a task, its affective state is selectively transmitted to the child under a policy-defined inheritance mask that marks each field as inherited, excluded, or attenuated, with the child's prior state blended in by a policy-defined function and the operation recorded in both lineage records. Affective inheritance is depth-limited, so that beyond a maximum delegation depth child agents instantiate with their own baseline. Across populations of interacting agents, affective contagion propagates through delegation inheritance, interaction exposure, and broadcast propagation, each governed by contagion damping factors strictly less than one, aggregate contagion limits, spiral detection, and quarantine escalation that suppresses outgoing contagion from volatile agents, together preventing runaway emotional spirals.
In an embodiment, the affective state field receives modulation inputs derived from biological signals produced by a human user. Raw signals are transformed by a feature extraction layer into abstract physiological state descriptors, which a policy-governed, confidence-gated coupling function maps to the modulation layer: user stress elevation may increase uncertainty sensitivity and risk sensitivity, fatigue may increase escalation tendency, and engagement may increase novelty appetite. The coupling is policy-bounded and preserves privacy, since raw biological data is not stored and the underlying physiological measurements are not persisted in the agent's lineage or memory field.
Disclosure Scope
This article describes the affective state of a semantic agent as a deterministic, policy-bounded control structure carried as a seventh schema field and consumed by deliberation and execution machinery, as disclosed in the cognition filing (U.S. Application No. 19/647,395). The disclosure encompasses the structured modulation layer of named control fields, each represented as a tuple of magnitude, decay rate, policy ceiling and floor, and timestamp; the enumerated modulation targets including promotion thresholds, search breadth, branch growth rates, decay rates for unpromoted candidates, escalation thresholds, persistence parameters, delegation routing preferences, and mutation acceptance thresholds; the deterministic update function driven by structured observations; the policy-bounded update mechanics with range bounds, rate limits, admissible triggers, update authority, and decay governance; the strict separation of concerns that prevents affective state from creating authority or relaxing policy; and the cross-primitive integration with confidence and forecasting.
The disclosure further encompasses the centralized, federated, decentralized, and embodied substrate embodiments; affective inheritance under scoped masks within delegation chains; the multi-agent contagion model with its anti-spiral governance bounds; and biological signal coupling under privacy-preserving, policy-governed coupling functions. The disclosure does not depend on any specific inference technology, language model, or execution runtime. It is defined by its structural and behavioral contract, and any implementation that maintains a deterministic, policy-bounded affective field consuming structured observations and modulating governed deliberation parameters instantiates the primitive, with domain-specific choices appearing only in the configuration of fields, update rules, decay parameters, and bounded modulation mappings.
