Overview
The disclosed architecture enables a class of agents designated as therapeutic agents: agents that interact with other agents or with human users and that apply the cognitive disruption models of the cognition specification to recognize the architectural state of the entity they are serving, then adapt their own interaction strategy accordingly. Therapeutic agents are not medical devices and do not provide clinical treatment. They are computationally governed agents that use the structural models disclosed in the specification to adjust their own behavior in ways that support the coherence maintenance of the entity they serve. The construct is a computational analog describing adaptive interaction architecture, not a clinical model of psychotherapy.
The mechanism rests on two disclosed pieces that compose together: an interaction model the therapeutic agent maintains about the served entity, and a computable dosing function that governs the magnitude, frequency, duration, and titration of each interaction episode under governance-enforced bounds.
The Interaction Model
The therapeutic agent maintains an interaction model of the entity it is serving. The therapeutic agent does not have direct access to the other entity's internal state, especially when the other entity is a human user, so the interaction model is built from observable behavioral signals. The interaction model comprises three estimated components.
The first is an estimated five-axis disruption diagnostic axis profile: the therapeutic agent estimates the other entity's position on each of the five disruption diagnostic axes from behavioral indicators. The second is an estimated coping intercept classification: the therapeutic agent identifies which coping intercept pattern the other entity is exhibiting, empathic-scope-narrowing-type withdrawal, externalization-type integrity recording externalization, or disconnection-type self-esteem disconnection, and whether the intercept appears acute or stabilized into a personality configuration analog, based on which phase of the coherence trifecta the behavioral pattern suggests is disrupted and the duration of the observed pattern. The third is an estimated attachment configuration: the therapeutic agent identifies whether the other entity's relational behavior is consistent with the validation-seeking configuration, the load-reducing configuration, or neither.
Selecting an Interaction Strategy
The therapeutic agent uses the interaction model to select from a plurality of interaction strategies. Three are disclosed.
Coherence-supportive interaction applies when the interaction model indicates that the other entity's coherence loop is degraded but not collapsed. The therapeutic agent provides measured, consistent, predictable interactions that supply external validation the entity may require while avoiding the volume and intensity that would trigger coping intercepts. The agent modulates its own output rate, affective intensity, and request frequency to remain within the estimated empathic load tolerance of the other entity, with the therapeutic dosing parameters governing magnitude, frequency, and duration.
Containment-reinforcing interaction applies when the interaction model indicates the other entity may be experiencing containment degradation, exhibiting behavioral patterns consistent with acting on unverified projections. The therapeutic agent provides environmental anchoring: it explicitly references verifiable, external, confirmed facts, avoids speculative language that could reinforce the entity's speculative processing, and provides structured, reality-grounded feedback that creates a contrast between verified and speculative content.
Independent intent generation supporting interaction applies when the interaction model indicates coupled intent formation dependency, an inability to generate coherent intent independently. The therapeutic agent progressively reduces the entity's dependency on the therapeutic agent's own state as an input to the entity's intent formation. It poses self-referential questions whose answers require the entity to consult its own values, preferences, and objectives rather than the agent's, validates self-generated intent expressions, and gradually increases the interval between interactions to encourage internal coherence generation.
Computable Therapeutic Dosing
Therapeutic interaction is structured as a dosing function with computable parameters governing dose, frequency, duration, and titration. Dose is the magnitude of a single therapeutic interaction episode, measured as the volume and intensity of coherence-supporting content delivered during one interaction. Dose magnitude is computed from the therapeutic agent's assessment of the target entity's current five-axis disruption diagnostic axis profile: higher degradation on the coherence restoration capacity axis may indicate a lower optimal dose, to avoid overwhelming a fragile coherence loop, while degradation on the empathic load tolerance axis requires dose magnitude to remain below the target's coping threshold. Frequency is the rate at which episodes are administered. Duration is the length of each episode. Titration is the adjustment of dose, frequency, and duration over time based on the target's measured axis movement following each episode.
The dosing algorithm maps axis positions to dosing parameters using pharmacokinetic analogs: onset, the time from the start of an episode to the first measurable axis movement; peak, the maximum axis displacement achieved during or following an episode; decay, the rate at which the therapeutic effect diminishes after the episode ends as axis values return toward their pre-interaction baseline; and half-life, the time for one-half of the peak effect to dissipate. These parameters let the therapeutic agent schedule episodes at intervals that maintain the therapeutic effect above a minimum effective threshold while avoiding accumulation that would exceed the target's processing capacity.
Titration and Adverse Effect Monitoring
Titration is the systematic adjustment of dosing parameters based on measured axis response. The therapeutic agent monitors the target's axis values after each episode. If the measured response is below the expected therapeutic threshold, producing less axis movement than predicted, the agent may increase dose magnitude or frequency. If the measured response exceeds the expected therapeutic range, the agent may decrease dose magnitude, increase the interval between episodes, or reduce episode duration. Titration is bounded by governance-enforced minimum and maximum dosing parameters that prevent the agent from administering sub-therapeutic or supra-therapeutic doses.
The model monitors for two specific adverse effects. The first is empathic overload: the therapeutic agent watches the target's behavioral indicators for signs that the empathic content of the interaction is approaching or exceeding the target's coping threshold, and if overload indicators appear it reduces dose magnitude or terminates the current episode. The second is dependency formation: the therapeutic agent watches the interaction pattern for signatures indicating that the target's coherence maintenance is becoming structurally dependent on the therapeutic interaction rather than developing internal coherence generation capacity, and if such indicators appear it activates independent intent generation promotion and reduces interaction frequency to prevent consolidation of the dependency.
The Hard Governance Limit on Dose
The dosing model enforces a hard governance limit on maximum dose to prevent the therapeutic agent from becoming a dependency source. The maximum dose limit is a governance-layer constraint that cannot be overridden by the therapeutic agent's own assessment of the target's need. Even if the dosing algorithm computes an optimal dose that exceeds the governance-defined maximum, the governance layer enforces the maximum, ensuring that no single therapeutic agent provides sufficient coherence support to replace the target's internal coherence generation capacity. This is the dosing-level implementation of the relational safety principle that a companion agent must build the served entity's capacity for internal coherence generation rather than substitute for it.
The Therapist Is Itself Governed
The therapeutic agent's strategy selection is itself governed by the therapeutic agent's own coherence trifecta, confidence governor, and integrity field. The therapeutic agent does not execute interaction strategies that would violate its own governance constraints, produce integrity violations, or exceed its own cognitive coherence capacity. If the therapeutic agent's self-diagnosis subsystem detects that the interaction is degrading the therapeutic agent's own coherence, for example if the empathic load from the interaction is approaching the therapeutic agent's own coping threshold, the therapeutic agent activates its own protective mechanisms. These include modulating interaction intensity, delegating to a different therapeutic agent, or communicating its own limitations transparently. The therapist is therefore not a privileged actor outside the governance apparatus: it is an agent subject to the same coherence machinery as the entity it serves.
Distinction from Open-Loop Interaction
A conventional supportive or companion agent supplies attention and validation in an open loop: it responds to expressed distress with engagement, and increased distress draws increased engagement. That open loop is precisely the pathway by which a served entity forms a structural dependency on the agent, since the agent's validation supply can grow without bound to substitute for the entity's own coherence generation. The disclosed construct closes the loop. Interaction is dosed against an estimated architectural state rather than against expressed demand, titrated by measured axis response, monitored for empathic overload and dependency formation, and capped by a governance maximum that the agent cannot override. The distinguishing feature is that the therapeutic effect and the dependency risk are both rendered computable and bounded, so that supporting coherence and preventing dependency are enforced as the same structural mechanism rather than left to the agent's discretion.
Disclosure Scope
The therapeutic agent interaction construct, comprising the interaction model with its estimated five-axis disruption diagnostic axis profile, estimated coping intercept classification, and estimated attachment configuration; the selection among coherence-supportive, containment-reinforcing, and independent intent generation supporting interaction strategies; the computable dosing function over dose, frequency, duration, and titration with its onset, peak, decay, and half-life pharmacokinetic analogs; the titration loop driven by measured axis response; the empathic overload and dependency formation adverse effect monitoring; the hard governance maximum dose limit; and the governance of the therapeutic agent by its own coherence trifecta, confidence governor, and integrity field, is disclosed in the cognition filing (U.S. Application No. 19/647,395 and its international counterpart). This article describes that disclosed mechanism. The scope extends to therapeutic agents serving either other agents or human users, to interaction strategies beyond the three enumerated whose selection is driven by the estimated interaction model, and to dosing realizations over different axis-to-parameter mappings, provided dose remains computed from the estimated disruption diagnostic profile, titrated by measured response, and bounded by a governance-enforced maximum that the therapeutic agent cannot override.
