Planning Graphs as First-Class Cognitive Structures

The semantic agent schema comprises structural fields, the intent field, context block, memory field, policy reference field, mutation descriptor field, lineage field, affective state field, and integrity field, that operate on verified state: each value reflects committed, auditable, governance-validated reality recorded in the agent's lineage. These fields give the agent no way to reason about hypothetical future states. The planning graph supplies that capacity. It is a mutable, memory-referenced, directed semantic structure representing one or more hypothetical future states of the agent, its environment, or both. It comprises a root node representing the agent's current verified state and a plurality of branches, each branch a distinct hypothetical trajectory: a sequence of speculative mutations, delegation outcomes, environmental transitions, or intent resolutions the agent is evaluating as a possible future.

A planning graph is not an execution plan, a schedule, or a commitment. It is a pre-execution construct that exists in a structurally distinct computational domain from the agent's verified execution memory. Each branch encodes a speculative mutation sequence, a projected outcome characterizing the expected terminal state if executed, an affective reinforcement tag encoding the emotional valence of the branch given the agent's current affective state and the projected outcome's alignment with intent, a trust slope projection encoding the trust slope trajectory that executing the branch would produce, a policy compatibility flag, and a branch classification label. Planning graphs are instantiated through defined interfaces, governed by policy, modulated by the affective state, constrained by the integrity field, and recorded in lineage only when promoted to execution.

Structural Separation from Verified Execution Memory

Planning graphs are maintained in structural separation from the agent's verified execution memory. The separation is not a software convention, a namespace distinction, or an access control policy; it is an architectural invariant enforced at the substrate level. The verified execution memory, comprising the committed values of all agent fields, the lineage of governance-validated mutations, and the accumulated results of executed operations, occupies a distinct computational domain from the planning graph structures. No mechanism permits a planning graph branch to modify verified execution memory directly without passing through the governance-validated promotion pathway.

The separation serves three purposes. First, speculative reasoning cannot contaminate verified state: an agent that constructs a branch projecting a successful outcome does not thereby acquire that outcome as verified memory. Second, the agent can maintain multiple contradictory hypothetical futures simultaneously, one branch projecting task success and another projecting task failure, without producing internal inconsistency, because both exist in the speculative domain. Third, the separation provides the structural basis for the containment layer. The boundary is enforced through a promotion interface: a governance-controlled gateway that receives a candidate branch, subjects it to the full governance evaluation pipeline (policy compliance, trust slope validation, integrity impact assessment, capability verification), and either admits it to verified execution memory as a committed mutation or returns it to the speculative domain with a rejection annotation. The promotion interface is the sole gateway, and its governance requirements are not waivable or bypassable by the agent's affective state, personality configuration, or operational urgency. The separation is bidirectional: when the forecasting engine constructs a planning graph it reads current verified state as the root node under snapshot isolation, so subsequent verified state changes do not automatically propagate into existing graphs.

The Forecasting Engine and Its Execution Cycle

The forecasting engine is a substrate module, instantiated at the agent level or zone level, responsible for constructing, evaluating, modulating, and managing planning graphs throughout their lifecycle. It is not an external service or a centralized scheduler; it operates on the agent's own state, subject to the agent's own policy, modulated by the agent's own affective and integrity fields. It comprises five principal components: planning graph instantiation logic, which creates graphs from current verified state by reading the intent field, context block, and memory field; an affective prioritization module, which orders and weights branches by the current affective state; a slope validation module, which evaluates each branch against the trust slope trajectory; a personality-based modulation filter, which adjusts construction and evaluation parameters from the personality field; and a pruning manager, which removes branches that are no longer viable.

The engine runs a defined execution cycle at each cognitive decision point, a synchronous component of the agent's deliberation pipeline rather than a background process. The cycle has six sequential phases. Initialization reads current verified state and constructs or refreshes the planning graph root node. Speculative mutation simulation applies each branch's hypothetical mutations to a sandboxed copy of the agent's state, never to verified execution memory, and computes the projected outcome; the simulation is deterministic, so identical input state and mutation sequence yield identical projected outcomes, and it is constrained by trust slope continuity and policy compatibility at every step rather than evaluated statistically over random rollouts. Slope projection and validation computes the hypothetical trust slope state for each simulated branch and confirms or flags continuity. Policy compatibility check evaluates each slope-eligible branch against current policy. Emotional reinforcement tagging assigns each remaining branch an affective reinforcement tag encoding alignment with the current affective disposition. Branch marking and pruning assigns each branch a classification label and schedules failed branches for removal.

Slope-Constrained Speculative Simulation

The forecasting engine's speculative simulation is slope-constrained: the trust slope trajectory serves as a structural filter determining which hypothetical futures the agent may evaluate for promotion. The constraint is a hard architectural boundary, not a soft preference or a ranking criterion. For each speculative branch the slope validation module computes a hypothetical Derived Anchor Hash (DAH') by applying the branch's speculative mutation sequence to a sandboxed copy of the agent's lineage and computing the trust slope hash that would result. The hypothetical DAH' is compared against the agent's current trust slope trajectory using the same continuity validation algorithm the governance infrastructure applies to committed mutations. If the hash chain relationship satisfies the cryptographic lineage requirements, the branch is slope-eligible; if the mutations would produce a lineage gap, a hash chain discontinuity, or a provenance violation, the branch is slope-ineligible.

The constraint operates prospectively, filtering branches before they reach the promotion interface, so the governance pipeline never receives a candidate that would fail trust slope validation. It interacts with the integrity field through an integrity impact projection: for each slope-eligible branch the integrity engine computes the projected change to the agent's integrity score across its three domains, and branches with negative integrity impact are not automatically disqualified but have their evaluation priority reduced. Only slope-eligible branches may be promoted. A slope-ineligible branch may be retained for introspective purposes, letting the agent understand why certain futures are structurally foreclosed, but it cannot advance through the promotion interface.

Branch Classification and the Containment Layer

Each branch is assigned one of four classification labels. An eligible branch has passed slope validation, satisfied policy compatibility, and received positive or neutral affective reinforcement; it is a viable promotion candidate, ranked by a composite score combining projected outcome quality, trust slope continuation magnitude, integrity impact projection, affective reinforcement strength, and intent alignment. An introspective branch is slope-eligible and policy-compatible but negatively reinforced; it is retained, not promoted, so the agent can examine its own aversions and detect affective bias. A delegable branch is slope-eligible and policy-compatible but better suited for transfer to a child agent. A pruned branch has failed slope validation or policy compatibility, exceeded entropy or compute thresholds, or been superseded. Classification is not permanent: an introspective branch may become eligible when the affective state shifts, an eligible branch may become pruned when conditions change, and the execution cycle re-evaluates classifications at each iteration.

The containment layer is a structural enforcement mechanism maintaining the separation between the speculative planning graph domain and verified execution memory. Every data element within a planning graph carries an immutable speculative marker that identifies it as non-verified content and cannot be removed or overridden by any operation within the planning graph domain; only the promotion interface, upon successful governance validation, strips the marker and re-tags content as verified. The layer enforces read isolation, so a query from a verified execution process receives the verified field value rather than a projected value from an active branch, and it prevents speculative content from being written to lineage as committed state. The layer defines a delusion boundary condition: a formally specified pathological state, containment collapse, in which speculative content is treated as verified reality and the agent can no longer distinguish what it has projected from what has actually occurred. The system provides containment integrity verification mechanisms, including periodic containment audits, boundary crossing monitors, lineage consistency checks, and behavioral coherence monitors, and on detection initiates a containment restoration protocol that suspends execution authority, quarantines affected structures, performs lineage forensic analysis, reconstructs verified state from the most recent governance-validated checkpoint, and re-initializes the containment layer.

Personality and Affective Modulation

The personality field is a structured data object of trait dimensions that deterministically shape the forecasting engine's instantiation logic, branch generation parameters, and evaluation criteria. It is not an aesthetic persona. Its trait dimensions include risk tolerance, which governs willingness to generate and promote high-variance branches; introspective depth, which governs resources allocated to introspective branch analysis; impulsivity, which governs the tendency to promote branches with reduced evaluation depth; fallback rigidity, which governs reversion to previously validated planning patterns; delegation preference, which governs the tendency to classify branches as delegable; and temporal planning horizon, which governs how far speculative projection extends into the future. The personality field may be set by static configuration, by policy-bound adaptation within policy-defined bounds, or by adaptive evolution through a feedback mechanism, with the configuration mechanism specified by the policy reference field and the evolution history recorded in lineage.

The affective state field modulates planning graph construction through defined coupling pathways, and is structurally distinct from personality modulation: personality encodes the slowly-evolving disposition, affect encodes the current, rapidly-changing orientation. Elevated risk sensitivity produces shallower branches with higher-confidence projections; elevated novelty appetite produces deeper branches exploring more distant futures. The affective state biases branch prioritization, influences delegation urgency through escalation-under-time-pressure and cooperation disposition values, and governs branch retention under failure through the persistence-under-partial-failure value. This emotional modulation shapes how the engine constructs and evaluates branches but does not determine whether branches are admissible for promotion: the governance requirements for promotion remain identical regardless of the agent's affective state.

The Executive Engine and Multi-Agent Aggregation

The executive engine is a substrate module that aggregates planning graphs from a plurality of agents operating within a shared operational scope, a zone, a delegation hierarchy, or a coordination group, into a unified executive graph representing the collective speculative state. It distinguishes two structural tiers: micro-planning graphs, the agent-level graphs each agent constructs from its own state, and the macro executive graph, the zone-level or group-level structure the executive engine builds by aggregating, aligning, and reconciling those micro-planning graphs. The executive graph is not a simple union; it is a synthesized structure that identifies inter-agent dependencies, resource contention points, and cooperative opportunities not visible from any single agent's perspective.

Aggregation proceeds by collecting each agent's active branches with their labels, affective reinforcement tags, slope projections, and policy compatibility flags; identifying branch intersections where branches from different agents reference the same resources, target the same delegation endpoints, or project mutually dependent outcomes; constructing executive graph nodes representing the coordinated state transitions those intersections require; and evaluating the result for global consistency against zone-level trust slope, aggregate resource budgets, and zone-level policy. The executive graph arbitrates among agents' branches using three criteria applied in priority order: slope compatibility first (combinations maintaining trust slope continuity at both agent and zone level are preferred), emotional reinforcement alignment second (combinations producing positive affective reinforcement for the majority of participating agents, subject to policy-defined participation thresholds), and personality profile alignment third (combinations consistent with each agent's personality field). The executive graph maintains its own containment layer, structurally separate from the individual agents' containment layers, so zone-level speculative coordination does not contaminate zone-level verified state.

Conflict Resolution and Execution Without Schedulers

When branch intersections produce conflicting demands, the executive engine runs a structured, deterministic conflict resolution protocol governed by policy-defined arbitration rules. Overlap detection classifies the conflict by type: resource contention, outcome contradiction, sequencing incompatibility, or delegation collision. Compatibility assessment determines whether the conflict can be resolved by branch modification (adjusting timing, resource allocation, or sequence without changing projected outcomes), requires suppression or rerouting, or is irreconcilable and must escalate to governance authorities. Arbitration, where modification fails, selects which branch takes precedence using the three-criteria ordering plus integrity impact, hierarchical position in the delegation hierarchy, and global impact assessment. When standard arbitration is inconclusive, an emotional quorum override acts as a tiebreaker: if a policy-defined supermajority of affected agents exhibit strong positive affective reinforcement toward one branch, that branch is promoted. The override operates within governance constraints, not over them. Personality-driven suppression or rerouting may replace a conflicting branch from an agent with low conflict tolerance or high fallback rigidity with an alternative from the same agent's graph rather than suppressing it outright.

This architecture replaces centralized scheduling with forecasting-driven branch promotion. Each agent constructs its own planning graph, evaluates its own branches through its own execution cycle, and an agent begins executing a task not because a scheduler assigned it but because the agent's own forecasting engine generated a branch representing the task, classified it eligible, and promoted it through the governance-validated promotion interface. Coordination emerges from the alignment and conflict resolution of independently generated plans rather than from a central authority imposing plans from above. The pattern eliminates the single point of failure (other agents continue if one engine fails), distributes the computational burden of planning, preserves agent autonomy, and supports heterogeneous agent populations with different personality configurations, capability envelopes, and policy constraints.

Dormancy, Reinterpretation, and Deferred Promotion

Each branch is temporally anchored with a timestamp recording the verified state it was generated from and a projection window specifying when its projections cease to be valid. The pruning manager enforces temporal expiration, slope invalidation, policy revocation, entropy threshold pruning, compute budget pruning, and mutation-triggered pruning, recording each pruning event in lineage as cognitive metadata. Beyond pruning, the engine supports branch dormancy: a branch that is currently non-promotable but retains potential future value is preserved in a reduced-resource state, excluded from active simulation, slope projection, and affective reinforcement cycles while consuming minimal resources. Temporal expiration still applies, so a dormant branch whose projection window expires is pruned even if never reactivated.

Branch reinterpretation re-evaluates a dormant or active branch under changed conditions and assigns it a new classification or projected outcome: an introspective branch may be reinterpreted as eligible when the affective state shifts, a slope-ineligible branch as slope-eligible when the trust slope trajectory changes, or a policy-incompatible branch as policy-compatible when policy is updated. Deferred promotion retains a branch that was not eligible when first evaluated and promotes it later when conditions render it eligible, so the agent is not required to make irrevocable planning decisions at the time of initial evaluation. The forecasting engine also feeds the confidence governor: when every branch in the active planning graph is pruned, introspective, or slope-ineligible, it transmits a negative viability signal that reduces the agent's confidence metric and shifts it into a non-executing cognitive mode (broader search, inquiry generation, delegation exploration) rather than acting without a viable plan.

Disclosure Scope

The planning graph as a mutable, memory-referenced, first-class cognitive structure structurally separated from verified execution memory; the promotion interface as the sole governance-controlled gateway between the speculative and verified domains; the forecasting engine, its five components, and its six-phase execution cycle; slope-constrained speculative simulation using the hypothetical Derived Anchor Hash (DAH'); the four-way branch classification taxonomy of eligible, introspective, delegable, and pruned; the containment layer and the delusion boundary condition with its restoration protocol; personality-field and affective-state modulation of planning behavior; the executive engine aggregating micro-planning graphs into a macro executive graph; conflict resolution with the emotional quorum override; branch dormancy, reinterpretation, and deferred promotion; and forecasting-driven branch promotion as the coordination primitive that replaces centralized scheduling, are disclosed in the cognition filing (U.S. Application No. 19/647,395) at Chapter 4. This article describes that disclosed mechanism using the specification's own terminology.

The scope extends to the centralized, federated, decentralized, and embodied deployment models disclosed for the forecasting and executive engines, across which the architectural properties (planning graphs as first-class structures, structural separation, the containment layer and delusion boundary, personality and emotional modulation, executive graph aggregation, and the forecasting execution cycle) are invariant. Implementation choices including the underlying simulation mechanism, the planning-graph storage format, and the arbitration scoring function are non-limiting, provided the implementation respects the containment layer, the trust slope constraint, and the governance-validated promotion interface.