1. The Disruption-Recovery Problem in Logistics

A freight network is a continuous stream of disruptions. A West Coast port congests and dwell times double. A winter storm closes an interstate corridor and reroutes every truck through a single mountain pass. A driver approaching the FMCSA eleven-hour driving limit cannot legally complete the last leg, and the Electronic Logging Device records the violation if dispatch pushes anyway. An ocean carrier blanks a sailing and a container that was the critical input to a downstream assembly line is now two weeks late. Each of these is foreseeable as a class even when the specific instance is not, and each demands a fast switch to an alternative plan that was, ideally, thought through before the disruption arrived.

Optimization-based logistics planning is organized differently. A transportation management system or a route optimizer computes a current best plan against current conditions, dispatches it, and when conditions change, re-optimizes from scratch. The alternatives the solver considered during optimization are intermediate state, released once a solution is selected. Recovery is therefore performed as a fresh solve under time pressure. The architecture described here is organized around the opposite retention policy: a portfolio of pre-evaluated contingencies, a structured record of which constraints bound the search, and a governed transition from the active plan to a prepared one.

The operation also sits inside a dense obligation stack that any switch must respect. Domestic trucking is bound by FMCSA Hours of Service rules and the ELD Mandate, which cap driver duty cycles and require auditable evidence of compliance. Ocean shipping is bound by IMO safety-management and cyber-risk-management obligations, including resolution MSC.428(98). Air cargo is bound by the IATA Operations Manual and the Dangerous Goods Regulations, whose deviations have air-safety consequences. The integration between enterprise planning and operational technology is described by ISA-95, its cybersecurity zoning by ISA-99/IEC 62443, and the integrity of the systems running the planning by the NIST Cybersecurity Framework. Defense and multimodal logistics add Joint All-Domain Command and Control coordination expectations across operational domains. A contingency that violates any of these is not a usable contingency, which is why the architecture described here gives a candidate plan a place to carry the reason it was rejected, distinguishing a compliance rejection from a cost ranking.

2. What the Operation Actually Requires

Read structurally rather than procedurally, the logistics obligation stack points past a single working plan that is rebuilt after each disruption. It points toward a planning substrate that maintains, simultaneously, the active dispatch plan, a portfolio of pre-evaluated alternative plans indexed by disruption class, and the constraint evaluation that establishes each alternative is admissible before it is offered to a dispatcher. In the described embodiments, switching from the active plan to a contingency is itself a recorded, credentialed decision, so that the question "why did the load reroute and was the substitute driver legal" has a structural answer rather than a reconstructed one.

This is exactly the shape the forecasting engine provides, and it is why the disclosed technology maps onto freight without being bent to fit. The engine maintains planning as planning graphs whose branches are alternative futures the agent may realize. Branches are classified, held under containment apart from execution, and scored, and the validated, highest-ranked branch is the one eligible to promote to execution. The match between this primitive and the logistics requirement is the subject of the rest of this article.

3. The Forecasting Engine, Applied to Freight

The Forecasting Engine is disclosed as Chapter 4 of United States Patent Application 19/647,395, where it is one cognitive domain of a semantic agent whose other domains (affective state, integrity, personality, confidence, and capability) govern how planning is shaped and when a plan may act. Several disclosed mechanisms carry directly into logistics.

Planning graphs as the unit of planning. Rather than a single optimized route set, the logistics agent maintains a planning graph in which the active dispatch plan and every contingency are branches of one structure. A weather-reroute branch, a capacity-substitution branch, a modal-shift branch (truck to rail, ocean to air for a hot order), and a driver-relay branch coexist in the graph, each carrying its own resource loading and its own constraint evaluation.

Branch classification. The disclosure classifies branches as eligible, introspective, delegable, or pruned. An eligible branch is a complete alternative the agent may choose to realize, ranked by a composite score; the highest-ranked eligible branch at any evaluation point is the agent's working forecast. An introspective branch is explored to inform the agent without being a candidate for execution, which in freight corresponds to stress-testing a what-if (what if the storm tracks north) without committing it. A delegable branch is transferred to a child planning graph, the mechanism by which a network-level plan hands a regional reroute to a regional agent. A pruned branch is scheduled for removal once superseded, but is retained for a defined period before deletion so that the planning history is recoverable. This classification is what lets contingencies remain dormant-but-ready rather than discarded.

The containment boundary. Branch expansion happens inside a containment boundary: speculative branches are structurally marked and do not commit state changes to verified agent state. Forecasting is non-executing. A reroute branch can be elaborated, costed, and checked against Hours of Service in full detail without a single load actually moving, because the boundary makes planning and acting different operations. This is the structural answer to the concern that an autonomous logistics planner might "decide" to reroute freight on its own; in the described architecture, speculation reaches execution only through the promotion gate.

Personality- and affect-modulated expansion. The disclosure modulates branch expansion by the agent's personality field (trait dimensions such as openness) and affective state field (uncertainty- and risk-related control values). In a freight deployment this is the tunable difference between a conservative planner that expands few, low-variance contingencies and an exploratory planner that maintains a wider portfolio under high uncertainty, expressed as governed agent parameters rather than ad hoc heuristics.

Slope-constrained, policy-checked branches. Branches pass a slope-validation and policy-compatibility check; only slope-eligible branches that survive the policy check may be promoted. In logistics terms, a contingency that would violate a duty-cycle limit, a dangerous-goods segregation rule, or an OT security zone boundary fails the policy check at evaluation time and is withheld from promotion. The compliance constraint is encoded as a branch-validation predicate, so a plan that fails the predicate is not offered to a dispatcher.

The promotion gate and confidence-gated dispatch. Promotion of a branch from contained speculation to active execution passes through a promotion gate, and dispatch to action is confidence-gated: a plan acts only when the agent's confidence determination clears threshold, otherwise the agent remains in a non-executing mode and continues planning. Switching to a contingency is therefore a credentialed, recorded promotion event, producing the auditable evidence that the ELD Mandate, IMO safety-management documentation, and contract carriers' own audit trails require, as a by-product of the decision rather than a later reconstruction.

Executive-graph arbitration across agents. When planning is distributed across mode-specialized or region-specialized agents, an executive engine composes their planning graphs through intersection detection and conflict resolution into a macro executive graph. When the ocean agent promotes a port-diversion branch that lands containers at a terminal the drayage agent has no capacity for, the arbitration surfaces the resource conflict before promotion rather than when trucks queue at the gate.

4. Deployment Options and Embodiments

The application is not a single instance. The same primitive supports a range of grounded deployments:

  • Over-the-road trucking. Driver-relay, team-conversion, and reroute branches held against Hours of Service and ELD constraints; promotion produces the duty-status evidence the audit requires.
  • Ocean and intermodal. Sailing-blank and port-diversion contingencies maintained against IMO MSC.428(98) safety-management and cyber-risk obligations, with drayage and rail handoff modeled as delegable branches to downstream agents.
  • Air cargo. Modal-shift-to-air branches for hot orders validated against IATA Operations Manual procedures and Dangerous Goods segregation rules before they are offered.
  • Plant- and warehouse-edge logistics. Planning agents that sit at the ISA-95 boundary between enterprise planning and operational technology, with branch promotion constrained to respect the ISA-99/IEC 62443 zone-and-conduit model and NIST CSF integrity controls on the systems doing the planning.
  • Multimodal and defense logistics. Cross-domain plans coordinated through executive-graph arbitration in line with Joint All-Domain Command and Control expectations, where contingencies must hold across modes and domains simultaneously.
  • Shadow-mode adoption. The forecasting engine runs alongside an existing transportation management system, ingesting its plans and building the contingency portfolio as a governed side effect, before any promotion authority is granted. Authority is staged: first advisory, then authoritative for contingency switching, then the planning substrate of record, with the legacy system becoming a downstream view.

Each embodiment reuses the same disclosed mechanisms: classified branches, a containment boundary, policy-checked promotion, confidence-gated dispatch, and executive-graph arbitration. The variation is in which obligation stack supplies the branch-validation predicates and which connectors feed the planning graph.

5. Why This Differs From Re-Optimization

Adding a scenarios feature to a route optimizer leaves the underlying shape unchanged, because the optimizer still admits one promoted plan at a time and treats alternatives as intermediate state. The forecasting engine changes the shape: alternatives are first-class, classified, retained, and individually compliance-validated, and the act of switching is a governed promotion rather than a re-solve. Speculation is structurally separated from execution, so the system can think about a hundred reroutes without moving a single load, and it can prove, after the fact, exactly which contingency was promoted, under what confidence, and against which constraints. That separation is the defensible core of the application.

Disclosure Scope

This article describes an application of the Forecasting Engine disclosed in United States Patent Application 19/647,395, and is offered as an enabling public disclosure of that application to logistics, freight, and supply chain planning. The forecasting mechanisms relied upon (planning graphs, branch classification into eligible, introspective, delegable, and pruned branches, the containment boundary with immutable speculative markers, personality- and affect-modulated branch expansion, slope-constrained and policy-checked branches, the promotion gate, confidence-gated dispatch, and executive-graph arbitration) are disclosed in that application. The regulatory and operational frameworks referenced (FMCSA Hours of Service, the ELD Mandate, IMO MSC.428(98), the IATA Operations Manual and Dangerous Goods Regulations, ISA-95, ISA-99/IEC 62443, the NIST Cybersecurity Framework, and Joint All-Domain Command and Control) are external standards described here only to situate the application; they are not claimed. Where this article describes deployment specifics, connector behavior, or adoption staging beyond the cited disclosure, those are illustrative embodiments and not limitations on the disclosed invention.