Mechanism
Emotional quarantine is a restricted execution mode that the system enters when an agent's affective state field exhibits volatility that may compromise the reliability of the agent's deliberation. The affective state field is a deterministic, policy-bounded modulation layer composed of named control fields, each encoding a dimension of the agent's dispositional orientation, such as uncertainty sensitivity, ambiguity tolerance, novelty appetite, persistence under partial failure, escalation under time pressure, risk sensitivity, and cooperation disposition. Each named control field is updated by a deterministic update function that draws on structured observations derived from execution outcomes and environmental signals, and each update is clamped to policy-defined bounds and recorded in the agent's lineage. The system monitors these fields for volatility, and when a volatility condition is detected the agent is routed to the emotional quarantine state.
Volatility is defined as a condition in which one or more named control fields exhibit rapid, high-magnitude oscillations, or in which the agent's composite affective deviation from baseline exceeds a policy-defined threshold. Quarantine is not a shutdown and it does not suppress the agent's affect: the agent continues to process observations and update its affective field. What quarantine does is reduce the agent's operational scope until its affective state stabilizes within governable bounds.
The Volatility Detector
The quarantine condition is assessed by a volatility detector that computes a composite volatility metric from the recent update history of all named control fields. In an embodiment, the metric is implemented as the sum of absolute update magnitudes across all fields within a sliding time window, normalized by the number of fields and the window duration. This treats volatility as a first-class measured property of the affective layer rather than as a derivative of confidence or error.
When the composite volatility metric exceeds the quarantine threshold, the agent transitions into quarantine. The detector operates over the same update history that is already recorded in lineage, so its inputs are the policy-bounded, audited affective mutations described above, not a separate or unaccountable signal stream.
Restrictions Imposed During Quarantine
An agent in emotional quarantine is subject to a defined set of restrictions, each of which constrains the agent's outbound effect without halting its internal cognition. The agent's promotion thresholds are elevated to their policy-defined maxima, preventing aggressive candidate advancement while the agent is in a volatile state. The agent's mutation acceptance thresholds are elevated, preventing the agent from incorporating externally proposed mutations, such as those proposed by an inference engine, while volatile. The agent's execution is routed through an additional validation layer that applies stricter admissibility criteria than the standard governance gate.
The agent's delegation authority is suspended. This restriction is specifically aimed at preventing the agent from propagating its volatile affective state to child agents. Because a parent agent's affective state is selectively transmitted to a child at delegation time under scoped inheritance rules, suspending delegation during quarantine closes the channel by which instability would otherwise spread down the delegation graph.
Hysteretic Recovery
Release from quarantine is governed by hysteresis. The agent transitions back to normal operation when the composite volatility metric falls below a recovery threshold, and the recovery threshold is set below the quarantine threshold. The separation between the two thresholds prevents oscillatory quarantine-release cycles in which a metric hovering near a single boundary would repeatedly enter and exit the restricted state. The specification describes this relationship qualitatively: the recovery threshold sits below the quarantine threshold to provide hysteresis. It does not fix the numeric magnitude of either threshold or of the gap between them, and those values are matters of policy configuration.
The lifecycle is therefore a state machine: normal operation feeds a volatility detector; when the composite metric exceeds the quarantine threshold the agent enters the quarantine state and its restricted mode; and when the metric falls below the recovery threshold the agent returns to normal operation. Throughout, the agent's affective field continues to be updated by the same deterministic, policy-bounded update function used in normal operation.
Quarantine as a Circuit Breaker
The quarantine mechanism functions as a circuit breaker for affective instability. Rapid affective change may arise from adversarial inputs, environmental turbulence, or cascading failure conditions, and an agent undergoing such change is at elevated risk of making consequential decisions on the basis of an unstable disposition. Quarantine prevents such an agent from making those decisions and from propagating its unstable modulation state to other agents, without discarding the agent's in-progress cognition. Because the agent keeps processing observations and updating its affective field during quarantine, it can stabilize and resume coherent operation as soon as the recovery condition is met, rather than requiring a cold restart.
Composition With Contagion Control
Quarantine composes with the system's affective contagion controls. In multi-agent operation, affective state propagates between agents through delegation inheritance, interaction exposure, and broadcast, and the system bounds this propagation to prevent runaway emotional spirals through mechanisms such as contagion damping, aggregate contagion limits, and spiral detection. Quarantine adds a further constraint: when an individual agent's volatility metric is elevated, the agent's outgoing contagion is suppressed, so its affective state is not transmitted to other agents through interaction exposure until the agent exits quarantine. This prevents a volatile agent from seeding instability across the broader agent population through the contagion channels, complementing the suspension of delegation authority that closes the inheritance channel.
Quarantine Does Not Override Governance
Quarantine operates within the strict separation of concerns that the affective layer maintains with respect to governance. Affective state, including the volatility that triggers quarantine, modulates how the agent deliberates, but it does not determine execution admissibility. Entering quarantine tightens the conditions under which the agent may act, but it does not grant the agent authority it lacks, validate or invalidate truth claims, or relax policy bounds. The restrictions quarantine imposes, raising promotion and mutation acceptance thresholds, suspending delegation, and routing through additional validation, all move in the direction of greater constraint and are themselves bounded by policy configuration. The governance gate continues to evaluate execution admissibility on its own criteria independently of the agent's affective state.
Disclosure Scope
Emotional quarantine and volatility management, comprising the detection of a volatility condition in which named control fields of the affective state field oscillate rapidly or the composite affective deviation from baseline exceeds a policy-defined threshold, the computation of a composite volatility metric over the recent update history within a sliding window, the routing of the agent into a restricted execution mode that elevates promotion thresholds to their maxima, suspends delegation authority, elevates mutation acceptance thresholds, and applies an additional validation layer, and the hysteretic recovery in which the recovery threshold sits below the quarantine threshold to prevent oscillatory quarantine-release cycles, together with the suppression of an agent's outgoing affective contagion while its volatility metric is elevated, is disclosed in the cognition filing (U.S. Application No. 19/647,395 and its international counterpart) in the affective state chapter. This article describes that disclosed mechanism. The scope extends to embodiments in which the composite volatility metric is computed by alternative normalizations over the recorded affective update history, provided the agent enters a policy-bounded restricted execution mode on exceeding a quarantine threshold and is released under a separate, lower recovery threshold, and provided the agent's affective field continues to be updated under the deterministic, policy-bounded update function during the restricted mode.
