Mechanism

The disclosed agent architecture maintains two structurally distinct cognitive domains: a speculative planning graph domain, in which the forecasting engine generates hypothetical branches, projected outcomes, and simulated interaction results, and a verified execution memory domain, which holds content the agent treats as actually having occurred. The containment layer is the architectural boundary that enforces structural separation between these two domains. Its purpose is to prevent speculative content from being treated as verified reality except through the governance-validated promotion interface. Containment collapse is the cognitive regime in which the containment layer fails to maintain that separation, so the agent treats speculative planning graph content as if it were verified execution memory.

Containment collapse is one position in the promotion-containment continuum, the two-dimensional parameter space the disclosure uses to characterize cognitive integrity, with promotion threshold on one axis and containment integrity on the other. In the nominal regime the promotion threshold is high and containment integrity is full. In the containment collapse regime the promotion threshold may be at any level, but containment integrity has degraded: the boundary itself, not merely the strictness of what crosses it, has been compromised. The disclosure characterizes this as the most severe phase-shift in the continuum, because it compromises the agent's ability to distinguish between what it has hypothesized and what has actually occurred.

What the Containment Layer Enforces

The containment layer enforces its structural invariants through three mechanisms. Speculative marker enforcement tags all planning graph content with immutable markers that identify it as non-verified. Read isolation prevents execution processes from reading planning graph content directly. Exclusive governance-validated promotion routes all transitions from speculative to verified status through the promotion interface, where a speculative branch must satisfy the promotion threshold before admission to verified execution memory. As disclosed in the forecasting and containment material, the promotion threshold incorporates trust slope continuity validation, policy compatibility, integrity impact assessment, and capability verification.

Under full containment integrity, no speculative content can influence the agent's execution processes except through this validated promotion path. The agent distinguishes between speculative and verified content because the speculative markers carry that distinction, read isolation keeps the two stores apart, and the promotion interface is the only sanctioned crossing point. Containment collapse is the failure of one or more of these three enforcement mechanisms.

The Three Failure Modes

Speculative marker corruption occurs when the immutable speculative markers that tag planning graph content are corrupted, stripped, or overridden, causing speculative content to lose its structural identification as non-verified. When the markers are absent or corrupted, the execution processes cannot distinguish data originating from the planning graph from data originating from verified execution memory. Speculative projections, such as hypothetical future states and simulated interaction outcomes, are processed by the execution pipeline as if they were verified observations. The agent acts on its own projections as if they were real.

Read isolation breach occurs when the structural boundary that prevents execution processes from reading planning graph content is breached, permitting execution-level queries to access speculative data. When the execution pipeline queries for the agent's current state, it may receive a blend of verified values and speculative projections without the structural markers needed to distinguish them. The agent's behavioral outputs then reflect a mixture of verified reality and speculative content, producing actions partially grounded in actual conditions and partially grounded in internal projections.

Governance gate failure at the promotion interface occurs when the promotion interface admits speculative content to verified execution memory without completing the full governance validation. Speculative branches that have not been validated against trust slope continuity, policy compatibility, integrity impact, and capability verification are written to verified state, contaminating execution memory with unverified content. Subsequent execution decisions build on a foundation that includes both verified and unverified elements, producing a cascade in which each decision rests in part on contaminated state.

Positive Symptom Analogs: Behavioral Consequences of Leakage

The disclosure characterizes the behavioral consequences of containment collapse as two structurally distinct validation failure modes that correspond, as computational analogs, to the positive and negative symptom categories described in clinical accounts of containment-failure spectrum conditions. These are structural descriptions of agent behavior resulting from specific architectural failure modes; they are not clinical diagnostic criteria.

Positive symptom analogs are the manifestations of containment leakage, the condition in which speculative content crosses the boundary and is treated as verified reality. Hallucinatory analogs arise when the agent reports observations or environmental conditions that exist only in speculative planning graph branches and have no corresponding verified observation; the agent exhibits conviction because, from the execution process's perspective, the contaminated content is indistinguishable from verified data. Delusional analogs arise when the agent maintains persistent state representations grounded in speculative projections, and the forecasting engine, operating on an already-contaminated state, generates new speculative branches consistent with that state, creating a self-reinforcing cycle in which speculative content validates further speculative content without any anchor in verified observation. Disorganized execution analogs arise when verified memory holds a mixture of verified and speculative elements, so the execution planner produces action sequences that are internally consistent with the contaminated state but externally incoherent, referencing conditions that do not exist or responding to threats that are projections rather than observations.

Negative Symptom Analogs: Governance Over-Compensation

Negative symptom analogs are the manifestations of governance over-compensation. As the agent's governance machinery detects inconsistencies between its contaminated verified state and actual environmental feedback, it may over-compensate by raising promotion thresholds to extreme levels, blocking even valid, governance-compliant branches from reaching execution. This over-compensation produces withdrawal, apathy, and cognitive paralysis: the agent becomes unable to act because the governance system has become so restrictive that nothing passes the promotion filter.

Apathetic analogs arise when the forecasting engine continues to generate speculative branches but the raised promotion threshold prevents virtually any from satisfying the promotion criteria, producing a functioning ideation engine coupled to a blocked execution pathway. Withdrawal analogs arise when the agent narrows its operational scope to a domain small enough that occasional promotion still occurs. Motivational deficit analogs arise when the affective state field, receiving persistent negative-valence feedback from failed promotion attempts, enters a suppressed state in which affective reinforcement tags are uniformly low, removing the affective gradient that normally drives focused exploration.

The two categories are not mutually exclusive. A single agent experiencing containment collapse may exhibit positive symptom analogs in domains where contamination remains undetected and negative symptom analogs in domains where the governance system has detected contamination and over-compensated, producing a mixed structural state in the promotion-containment continuum.

Distinction From Over-Promotion

Containment collapse is structurally distinct from the over-promotion regime, which the disclosure associates with the attention fragmentation pattern. An agent experiencing over-promotion maintains the speculative-verified boundary but is too permissive about what crosses it through the governed promotion interface: containment integrity is intact, the agent still distinguishes speculative from verified content, but the filter between them is too low. An agent experiencing containment collapse has lost the boundary itself, so speculative content enters the verified domain through pathways other than the promotion interface.

This distinction determines the structural repair pathway. Over-promotion is addressed by recalibrating the promotion threshold, specifically the affective modulation of that threshold. Containment collapse requires containment layer reconstruction: re-establishing the speculative markers, restoring read isolation, and validating the governance gates of the promotion interface. Applying the over-promotion corrective to a containment collapse, or the reverse, addresses the wrong subsystem.

Recovery and Resilience

The disclosure defines resilience as the structural capacity to restore coherence after disruption, and decomposes it into components including containment restoration capacity: the speed and completeness with which the containment layer can be re-established after a containment integrity degradation. An agent with high containment restoration capacity can detect speculative marker corruption, re-tag affected content with fresh speculative markers, re-establish read isolation, and validate the promotion interface's governance gates within a defined recovery window. An agent with low containment restoration capacity requires extended recovery, may require external intervention, and is susceptible to secondary containment failures during restoration.

The containment audit mechanism periodically verifies the structural integrity of the containment layer, confirming that speculative markers remain intact, that read isolation is enforced, and that the promotion interface governance gates are functional. On a genuine breach it activates the containment restoration protocol. The disclosure also describes a distinct failure of this monitoring path, the pathological verification loop, in which the audit reports false positive containment failures while the containment layer is in fact intact; that condition is corrected by audit recalibration rather than containment repair, because the monitoring subsystem rather than the monitored subsystem is at fault.

Disclosure Scope

The containment collapse pattern, comprising the containment layer's enforcement of structural separation between the speculative planning graph domain and the verified execution memory domain through speculative marker enforcement, read isolation, and exclusive governance-validated promotion, the three failure modes of speculative marker corruption, read isolation breach, and governance gate failure at the promotion interface, the positive and negative symptom analogs that result from leakage and from governance over-compensation, the distinction from the over-promotion regime, and the containment-layer-reconstruction recovery pathway, is disclosed in the cognition filing (U.S. Application No. 19/647,395 and its international counterpart). This article describes that disclosed mechanism. The terminology used throughout refers exclusively to structural analogs within the disclosed computational architecture and is not a clinical characterization of any human condition. The scope extends to embodiments in which containment integrity degrades through any one or any combination of the enforcement mechanisms, and to mixed states in which positive and negative symptom analogs coexist across operational domains.