Mechanism
Temporal anchoring is the property that ties each planning graph branch to a point in the agent's verified history. In accordance with the disclosed embodiment, each planning graph branch is temporally anchored: it is associated with a timestamp recording the verified state from which the branch was generated, and a projection window specifying the temporal range over which the branch's speculative projections are considered valid. The anchor is not a free-standing time service or an externally authenticated time claim. It is a binding between a speculative branch and the verified state snapshot that produced it, together with the interval during which that branch's projections may still be trusted.
The purpose of temporal anchoring is to keep speculative reasoning from outliving its own assumptions. A planning graph branch is a hypothetical trajectory built from a sandboxed copy of the agent's state at a particular moment. As the agent's verified state and environmental conditions evolve, the assumptions that informed a branch's construction can become stale. Temporal anchoring ensures that branches do not persist indefinitely, consuming computational resources and potentially projecting from a verified state that no longer reflects reality.
The Projection Window
The projection window is the temporal scope of a branch's validity. When the current time exceeds the timestamp at which a branch's speculative projections cease to be considered valid, the branch is automatically reclassified as pruned. This is temporal expiration: the window closes, and the branch is no longer eligible for evaluation or promotion.
The duration of the projection window is not a fixed constant. It is determined by the agent's personality field, specifically the temporal planning horizon trait, in combination with the policy configuration. The temporal planning horizon is a scalar value encoding the depth of the agent's speculative projection into the future. An agent with an elevated temporal planning horizon generates branches with longer mutation sequences that project further forward, while an agent with a suppressed horizon focuses on near-term projections with shorter mutation sequences. The projection window follows from this disposition, so the same anchoring mechanism produces different lifecycles for agents configured for strategic versus tactical reasoning.
Pruning Within a Lifecycle
Temporal expiration is one of several pruning criteria that the pruning manager enforces to govern the lifecycle of planning graph branches. The criteria operate together. Slope invalidation prunes a branch when the agent's verified state evolves in a manner that invalidates the branch's slope projection, for example when a committed mutation changes the trust slope trajectory in a way that renders the branch's hypothetical Derived Anchor Hash discontinuous. Policy revocation prunes a branch when a change to the policy configuration renders the branch's speculative mutations inadmissible. Entropy threshold pruning removes the lowest-scoring branches when the number of active branches exceeds a policy-defined entropy threshold, preventing unbounded planning graph expansion. Compute budget pruning removes branches when the cumulative cost of maintaining the active planning graph exceeds a policy-defined compute budget. Mutation-triggered pruning evaluates all active branches for continued viability when the verified state undergoes a significant mutation, such as a major delegation event, environmental change, or policy update, and prunes those whose root assumptions the mutation has invalidated.
Slope invalidation pruning is evaluated at each forecasting execution cycle, and the compute and entropy criteria prioritize retaining eligible and introspective branches over delegable and low-scoring branches. The temporal anchor sits alongside these criteria as the time-based boundary of the same lifecycle, so a branch can leave the active set either because time has passed or because the verified state, policy, entropy, or compute conditions have changed.
Pruning Events as Lineage Metadata
The pruning manager records pruning events in the agent's lineage as cognitive metadata. Each record captures which branches were pruned, the pruning criterion that triggered removal, and the branch's evaluation state at the time of pruning. A temporally expired branch is recorded as such, distinct from a branch removed by slope invalidation, policy revocation, or budget pressure.
This metadata serves two functions. It enables post-hoc analysis of the agent's planning behavior, allowing an investigator to reconstruct why a branch left the active set and under what conditions. It also supports the forecasting-as-input-to-confidence mechanism, in which the forecasting engine provides the confidence governor with an aggregate viability assessment of the current planning graph. A record of which branches expired, and when, is part of the structured account of whether the agent's speculative reasoning has identified viable paths forward.
Anchoring and Branch Dormancy
Temporal anchoring interacts with branch dormancy, the state in which a branch is neither actively evaluated nor pruned but is preserved in a reduced-resource format for potential future reactivation. A branch enters dormancy when it is currently non-promotable but the pruning manager determines it has potential future value, when its evaluation score has fallen below the active evaluation threshold while remaining above the pruning threshold, or when the forecasting engine explicitly marks it dormant in response to environmental uncertainty. A dormant branch's speculative content, projected outcome, and evaluation metadata are preserved, but the branch is excluded from active simulation, slope projection, and affective reinforcement cycles.
The temporal anchor does not lapse when a branch becomes dormant. The pruning manager continues to apply temporal expiration to dormant branches: a dormant branch whose projection window expires is pruned even if it has not been reactivated. This prevents stale projections from accumulating in the dormant store and ensures that the time boundary established when the branch was anchored remains in force regardless of the branch's activity state.
Anchoring Across Reinterpretation and Deferred Promotion
The forecasting engine supports reinterpretation, in which a dormant or active branch is re-evaluated under changed conditions and assigned a new classification or projected outcome, and deferred promotion, in which a branch that was not eligible when first evaluated is retained and subsequently promoted when conditions change to render it eligible. Deferred promotion is how the engine implements temporal flexibility in speculative reasoning: the agent is not required to make irrevocable planning decisions at the time of initial evaluation, but may construct branches, evaluate them under current conditions, retain the most promising across state changes, and promote them when execution conditions are met.
Temporal anchoring is the boundary condition that makes this flexibility safe. A branch can be retained and deferred only while its projection window remains open. Once the window expires, temporal expiration prunes the branch regardless of whether reinterpretation or deferred promotion might otherwise have rescued it. The combination of dormancy, reinterpretation, and deferred promotion lets the engine manage branches over horizons that exceed any single evaluation cycle, and the temporal anchor bounds that horizon, so long-duration planning in uncertain environments does not become indefinite retention of assumptions that have already gone stale.
Distinction from Conventional Plan Caching
Conventional planning and caching systems retain plans or computed results and invalidate them on an ad hoc basis, often by a fixed expiry or by explicit cache eviction. Temporal anchoring differs in that the time binding is a structural property of each branch tied to the specific verified state snapshot the branch was generated from, and the projection window is derived from the agent's own temporal planning horizon rather than a global constant. Expiration is one criterion within a unified pruning lifecycle that also responds to trust slope continuity, policy changes, entropy pressure, and compute budget, and every expiration is recorded in the agent's lineage as auditable cognitive metadata. The result is that a speculative branch carries, as part of its structure, both the moment it was rooted in verified reality and the interval over which it may still be acted upon.
Disclosure Scope
Temporal anchoring of planning graph branches, comprising the association of each branch with a timestamp recording the verified state from which it was generated and a projection window specifying the temporal range over which its projections are considered valid, the determination of projection window duration from the agent's temporal planning horizon trait and policy configuration, temporal expiration as a pruning criterion operating alongside slope invalidation, policy revocation, entropy threshold, compute budget, and mutation-triggered pruning, the continued application of temporal expiration to dormant branches, and the recording of pruning events in the agent's lineage as cognitive metadata, is disclosed in the cognition filing (U.S. Application No. 19/647,395 and its international counterpart) at Section 4.13. This article describes that disclosed mechanism. The scope extends to embodiments in which the projection window is computed under different personality and policy configurations, and to deployments in which temporally anchored branches participate in dormancy, reinterpretation, and deferred promotion, provided each branch remains bound to the verified state from which it was generated and to the interval over which its projections are considered valid.
