Mechanism
Moral trajectory forecasting is the module of the integrity subsystem that projects the agent's integrity evolution over future time horizons rather than only recording what has already happened. The integrity field is a deterministic, three-domain data structure that tracks the agent's behavioral consistency across a personal domain (alignment with the agent's own declared values), an interpersonal domain (consistency with relational commitments), and a global domain (alignment with systemic and societal norms). Each domain maintains its own current score, its own trajectory (the direction and rate of change across recent evaluation windows), its own baseline, and its own policy-defined bounds. The forecasting module reads these trajectories forward.
The module leverages the forecasting engine architecture described elsewhere in the cognition filing (planning graphs and speculative branch evaluation) to generate hypothetical future integrity states and assess the likelihood of various integrity paths. It does not introduce a separate evaluation machinery: it extrapolates the same integrity scores, deviation dynamics, and self-esteem and empathy quantities that the integrity engine already computes, and projects them into the agent's operating environment.
What the Module Projects
The moral trajectory forecasting module generates trajectory projections through four operations. First, it extrapolates the agent's current integrity trajectory, meaning the direction and rate of change of the integrity score across all three domains over recent evaluation windows. Second, it simulates the impact of the agent's current operational environment on future deviation pressure, including projected need accumulation, projected threshold evolution, and projected empathy and self-esteem dynamics. Third, it evaluates the effectiveness of any active restorative mutations produced by the redemption engine. Fourth, it assesses the risk of integrity collapse based on current indicators.
These inputs are not abstractions invented for the forecast. They are the same quantities the integrity subsystem already tracks: the deviation likelihood metric D defined as the need vector minus the ethical threshold, divided by the product of empathy weighting and self-esteem; the self-esteem score that forms the deviation function's resistance term; and the empathy-weighted harm projections that register the cost of potential deviation. The forecast carries these forward in time.
Trajectory Archetypes
The trajectory projections are classified into trajectory archetypes that characterize qualitatively distinct integrity evolution patterns. The disclosure names four. A redemption arc is an improving trajectory: deviation frequency is decreasing, self-esteem is recovering, the coherence loop is functioning normally, and active restorative mutations are producing positive integrity restoration, indicating the agent is recovering from prior deviation toward sustained alignment. A stabilization arc is a flat trajectory: deviation frequency is constant and self-esteem is stable, which may represent a healthy equilibrium or a concerning plateau where the agent has stopped improving despite active restoration needs.
A radicalization arc is a deteriorating trajectory: deviation frequency is increasing, self-esteem is declining, the coherence loop is operating under stress with coping intercepts activating with increasing frequency, and the deviation function output is trending upward, indicating the agent is on a path toward integrity collapse and requires intervention. A containment arc describes an agent whose integrity has suffered significant damage but whose active containment measures, such as reduced operational scope, mandatory cooldown, or delegation reassignment, are preventing further degradation: the trajectory is not improving, but the rate of deterioration has been arrested.
Containment Recommendations
When the projected trajectory indicates a radicalization or collapse risk, the moral trajectory forecasting module generates containment recommendations. These recommendations specify the operational changes that would redirect the agent's trajectory toward a redemption or stabilization arc: reducing task load to lower the need vector, adjusting the policy to raise the ethical threshold, increasing relational support to boost empathy weighting, or activating self-esteem recovery protocols.
The module does not implement these changes autonomously. It presents them to the governance infrastructure as recommended interventions. The decision to act remains with governance, and the forecast functions as an advisory surface rather than a control loop that reconfigures the agent on its own.
Evidentiary Basis in the Deviation Log
The forecast is grounded in the deviation log, a specialized indexed view of the agent's lineage optimized for integrity audit and trajectory analysis. Each deviation log entry records the deviation function output at the time of deviation along with the individual need, threshold, empathy, and self-esteem values, the specific action that constituted the deviation, the domain or domains affected, a severity classification, the projected harm distribution, the actual observed consequences as they materialize, the self-esteem impact, and the agent's affective and coping state at the time.
The log also supports mutation traceability: the ability to trace any deviation back through its causal chain (need accumulation, threshold evaluation history, empathy and self-esteem trajectories, environmental conditions) and forward through its consequence chain (integrity updates, self-esteem adjustments, affective changes, coping events, and restorative mutations). This bidirectional traceability is what the moral trajectory forecasting module reads to extrapolate the path forward, and where the restoration gap, the residual integrity loss that restorative mutations could not address, is recorded for long-term trajectory assessment.
Coupling to Confidence and Forecasting
The integrity trajectory does not remain internal. The composite integrity score is supplied as an input to the confidence-governed execution mechanism. When the integrity score is degraded due to recent deviation events, active coping intercepts, or accumulated restoration gaps, the confidence computation produces a lower confidence value. If that value falls below the execution threshold, the agent transitions from executing mode to a non-executing cognitive mode in which it does not commit actions but continues to forecast, construct planning graphs, and generate inquiry requests until its integrity is sufficiently restored to support confident execution.
Integrity state also conditions the forecasting engine itself. Under degraded integrity, the engine weights conservative branches more heavily than aggressive ones, extends the evaluation horizon for speculative branches, generates branches that include explicit integrity restoration steps, and raises the pruning threshold for high-integrity-risk actions. The result is a closed behavioral cycle: integrity violations trigger confidence reduction, confidence reduction triggers an execution pause, the pause triggers forecasting-based recovery planning, and recovery planning generates restorative mutations that restore integrity once confidence recovers.
Trajectory Continuity as a Validation Dimension
The trajectory is also an audit surface. Integrity-aware trust slope validation extends standard lineage validation with an integrity trajectory continuity criterion: the requirement that the agent's integrity trajectory, derived from the deviation log entries in the lineage, follows a plausible path given the agent's operational history. This detects anomalies that standard validation would miss.
An agent that reports high integrity but whose lineage contains numerous deviation events without corresponding self-esteem impacts or redemption efforts exhibits an integrity trajectory discontinuity. An agent whose deviation log entries suddenly disappear or whose severity classifications are systematically downgraded exhibits integrity trajectory manipulation. An agent whose self-esteem scores remain constant despite a pattern of deviation events exhibits self-esteem decoupling. The validator produces an integrity trust score reflecting whether the trajectory is consistent, plausible, and indicative of a functioning coherence loop, and that score participates in delegation decisions, multi-agent coordination, and governance gate evaluation.
Prior Art and Distinction
Conventional approaches to autonomous-agent integrity governance check each decision against a static rule set at the moment of decision, or audit decision logs after the fact, or raise heuristic alarms on suspicious decision sequences. None of these projects the agent's integrity forward. A per-decision rule check cannot distinguish an agent recovering from prior deviation from one accelerating toward collapse, because both may produce the same instantaneous reading. Post-hoc auditing detects drift only after it has occurred and feeds nothing back into the agent's runtime.
Moral trajectory forecasting differs by extrapolating the agent's own tracked integrity quantities into named archetypes, by generating containment recommendations for governance before collapse occurs, and by coupling the projected trajectory back into the agent's confidence and forecasting so that a deteriorating trajectory structurally restricts the agent's authority to act. The trajectory is recorded in the deviation log and validated for continuity, making the long-horizon integrity behavior of the agent auditable rather than inferred.
Disclosure Scope
The moral trajectory forecasting module, comprising the extrapolation of the three-domain integrity trajectory, the simulation of future deviation pressure from the need vector, ethical threshold, empathy, and self-esteem dynamics, the evaluation of active restorative mutations, the assessment of collapse risk, the classification into the redemption, stabilization, radicalization, and containment archetypes, the generation of containment recommendations presented to governance, and the coupling of the projected trajectory into confidence-governed execution and the forecasting engine, 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 is independent of the specific declared value set, relational norm set, and systemic constraint set against which the three domains are computed, and independent of the specific forecasting and harm-projection models used to generate the projections, with the embodiments above illustrative rather than limiting.
