1. Vendor and Product Reality

Anduril Industries, founded in 2017 by Palmer Luckey and Trae Stephens with engineering leadership drawn from Palantir and SpaceX, has become the reference software-first defense prime. The company manufactures hardware, Ghost and Bolt unmanned aerial systems, Sentry surveillance towers, Dive-LD undersea vehicles, Roadrunner and Pulsar counter-air interceptors, the Fury collaborative combat aircraft, but the binding asset is Lattice, the open mission-autonomy software platform that fuses sensor data, maintains a common operating picture, and coordinates assets across heterogeneous fleets. Lattice runs on Department of Defense networks under JADC2-aligned programs, on Customs and Border Protection border-surveillance deployments, on United Kingdom and Australian defense programs, and increasingly on allied procurement across Europe and the Pacific.

Lattice solves a genuine integration problem in defense operations. Sensors, effectors, and command elements from different manufacturers, different services, and different classification levels need to share information and coordinate action. Lattice provides the data fusion and command integration layer that makes cross-domain autonomous coordination possible. The platform handles sensor data ingestion, object tracking, threat classification, asset allocation across heterogeneous systems, and the operator-facing display surface through which human commanders maintain situational awareness and authorize action. The mesh networking, edge compute, and identity layers are battle-realistic. Anduril's engineering culture and procurement-cycle discipline have produced a product that the analyst community broadly accepts as the modern reference for tactical mission autonomy.

Mission planning within Lattice generates courses of action based on available assets, threat assessments, rules of engagement, and operational objectives. The system evaluates options, recommends actions to human operators, and supports human-on-the-loop authorization for kinetic decisions. This is appropriate and reflects responsible engineering. Anduril publicly emphasizes the human-on-the-loop and human-in-the-loop posture, which both U.S. DoD policy and allied frameworks require for lethal autonomy. Within its scope, Lattice is rigorous, mission-defensible, and procurement-ready.

2. The Architectural Gap

The structural property Lattice's planning layer does not exhibit is governed speculative containment with branch classification and promotion thresholds as first-class cognitive constructs. Course-of-action generation produces options and evaluates them against criteria; cognitive forecasting maintains those options as speculative branches with independent state, evolving consequences, and structural containment that prevents speculative reasoning from contaminating the active mission plan. The distinction matters when plans must adapt to evolving situations under time pressure.

A defense system that generates three courses of action and selects the best one is performing optimization. A system that maintains all three as live speculative branches, continuously updating their projected outcomes as the situation evolves, classifying each by risk profile and time-to-commitment, and structurally containing them so that branch evaluation does not affect the currently executing plan, is performing cognitive forecasting. The second system can switch plans faster and with better information because the alternative plans have been maturing in parallel. In defense operations, the difference is measured in the quality of decisions under time pressure. When the situation changes rapidly, a threat emerges, an asset is lost, intelligence updates, a system with contained speculative branches already has alternative plans at varying stages of maturity. A system that must regenerate courses of action from scratch faces computational and cognitive delay at exactly the moment speed matters most.

Containment is not a convenience. In defense applications, it is a safety requirement. Speculative reasoning about engagement options must be structurally isolated from the active engagement plan. A system that allows speculative branch evaluation to influence current asset positioning, sensor allocation, or weapons-release authorization before the branch is formally promoted creates unpredictable behavior that degrades operational trust and creates a surface for adversarial deception, feed the system a plausible alternative scenario and watch its current behavior shift. The containment boundary is what makes speculation safe.

Branch classification matters equally. A branch that imagines a flanking maneuver but fails feasibility must be structurally distinguished from one that has cleared feasibility and policy checks and is eligible to advance, and from one that has been marked for delegation to a subordinate element. Each class exists in the planning graph with its own advancement rules and operator-facing semantics. Without classification, speculative branches look the same to the system, and the operator cannot distinguish a validated alternative from raw speculation, which collapses the value of carrying alternatives at all.

The distinction here is architectural rather than a matter of feature depth. Forecasting in this sense is a planning substrate, not a planner feature. Adding more course-of-action generators, better optimizers, or richer scenario libraries deepens a generate-and-select pipeline; it does not by itself introduce a containment boundary with immutable speculative markers, a branch classification taxonomy, and a promotion gate that content must clear before it can be treated as verified and acted upon. Those are properties of the substrate a planning layer runs on. This is not a claim about what Anduril's engineers could or could not build; it is a claim about which architectural layer the property lives in.

3. What the AQ Forecasting-Engine Primitive Provides

The Forecasting Engine treats the planning graph as a first-class cognitive structure: speculative branches held under structural containment, branch classification by maturity, a governed promotion gate, and an executive-graph aggregation property that records the cognitive history of planning. In this framing, planning is not reduced to optimization; it is the disciplined cultivation of alternatives under containment, with promotion as the only path from speculation to action.

Containment is enforced at the substrate level. In the disclosed architecture, planning-graph content generated during forecasting is tagged with an immutable speculative marker at the time of generation, and a containment layer defines a delusion boundary: speculative content has read access to state but cannot be written into execution memory as verified while the marker is intact. A speculative branch is a region of the planning graph that can read the active situational state but cannot write to the active plan or to the assets executing it. Only the promotion interface, upon successful governance validation, strips the speculative marker and retags the content as verified. Failure of the containment layer (a corrupted or improperly stripped marker) is treated as the architectural analog of delusion and is detected and re-initialized rather than allowed to propagate. This isolation is a structural property of the substrate, not a coding convention.

Branch classification is governed by a defined taxonomy. In the disclosed forecasting cycle, each candidate branch is evaluated for feasibility and policy compatibility and then marked as eligible (cleared for promotion), introspective (retained for further analysis rather than acted upon), delegable (suitable for transfer to a subordinate planning graph), or pruned (failed validation and scheduled for removal). Branches also move through a lifecycle of active, dormant, reinterpretation, and deferred promotion states. Each class and state carries different advancement rules and operator-facing semantics. Reclassification follows changing conditions: a branch whose supporting assumption is invalidated can be downgraded or pruned, and pruned branches and lifecycle transitions are archived in a cognitive history store rather than silently discarded, which supports forensic reconstruction.

Promotion is the load-bearing structural mechanism. Planning graphs pass through a promotion gate governed by a promotion threshold, and only eligible branches may advance. In the disclosed system the threshold is a governance check composed of feasibility, confidence, and policy compatibility; in a defense composition it can additionally compose rules-of-engagement compliance, asset availability, and authority-credentialed approval where required, so that promotion of a plan to execution can be conditioned on the human-on-the-loop authorization that defense doctrine demands. The substrate makes that authorization a structural gate rather than a display convention. The executive graph aggregates planning graphs across agents through intersection detection and conflict resolution and records the cognitive history of why a branch was promoted and when an option was first considered, which supports after-action review.

The primitive is technology-neutral with respect to the branch generator that seeds the graph: the course-of-action source may be classical search, a learned planner, a hybrid, or a scenario library, because containment, classification, and promotion act on the branches regardless of how they were produced. It composes hierarchically, so squad-level planning graphs can roll up into higher-echelon graphs under the same containment shape, and it composes with delegation, where a delegable branch transfers to a child planning graph with its speculative marker preserved. A skilled implementer could build this approach: represent the planning graph as a data structure whose speculative regions carry an immutable marker; enforce, at the memory layer, that marked content cannot be written to execution state; run a per-branch evaluation cycle that assigns eligible, introspective, delegable, or pruned status; gate promotion on a composed threshold that strips the marker only after validation; and journal transitions to a history store. The inventive step, the Forecasting Engine disclosed in United States Patent Application 19/647,395, is this contained forecasting substrate: speculative branches under an immutable-marker containment boundary, classified branch maturity, a governed promotion gate, and executive-graph aggregation as a condition for governed autonomous planning.

4. Composition Pathway

Anduril integrates with AQ as a domain-specialized mission-autonomy and effects-coordination surface running over the forecasting-engine substrate. What stays at Anduril: the Lattice data-fusion layer, the asset SDK, the sensor and effector integrations, the common-operating-picture display, the mesh-networking and edge-compute infrastructure, the hardware platforms, and the entire DoD and allied procurement relationship. Anduril's investment in defense-specific knowledge, JADC2 mappings, classification handling, ROE encodings, allied-interoperability protocols, remains its differentiated layer.

What moves to AQ as substrate: the planning layer becomes a forecasting graph rather than a course-of-action generator. Course-of-action generation continues to seed the graph with candidate branches; the substrate then maintains those branches under containment, classifies them, evolves their projected outcomes as situational state changes, and surfaces their maturity to the operator through the existing Lattice display semantics. Promotion of a branch to active requires the substrate-level threshold including the human-on-the-loop signature, which is wired into the existing Lattice authorization flow rather than replacing it.

Integration points are well-defined. Lattice sensor fusion feeds situational state into the substrate as credentialed observations; intelligence updates trigger automatic reclassification of affected branches; asset-availability changes prune or downgrade branches that depended on the lost asset. The executive-graph aggregation feeds Anduril's existing after-action review tools as credentialed lineage rather than as a separate log. Cross-echelon planning composes in the same primitive shape, a squad-level planning graph is a sub-graph of the platoon graph under the same containment rules, which allows allied interoperability to be expressed as planning-graph federation rather than as ad-hoc liaison.

The new commercial surface is forecasting-as-substrate for defense customers who need auditable evidence that the autonomous-planning systems coordinating their assets maintained governed speculative containment. The substrate belongs to the customer's authority taxonomy and is portable across hardware refreshes and software upgrades. Lattice's fusion and effector value remains the differentiated layer through which a customer accesses that substrate.

5. Commercial and Licensing Implication

The fitting arrangement is an embedded substrate license: Anduril embeds the AQ forecasting-engine primitive into Lattice and sub-licenses planning-graph participation to its defense customers as part of the platform subscription or program-of-record line. Pricing is per-credentialed-planning-graph or per-promotion-event rather than per-seat, which aligns with how defense customers actually consume autonomy.

What Anduril gains: a structural answer to a recurring question in defense-autonomy oversight, whether autonomous-planning systems are governed planners or optimizers with an authorization step; a defensible position relative to adjacent command-and-control software such as Palantir's Maven Smart System and to primes' homegrown stacks, by raising the architectural floor to contained, auditable speculation; and a posture aligned with the direction of U.S. autonomy policy, including DoD Directive 3000.09 on autonomy in weapon systems, which emphasizes appropriate human judgment over the use of force. What customers gain: auditable planning-graph lineage, portable across mission and theater, expressible in a single primitive shape across echelons. The framing is deliberately limited: the Forecasting Engine does not replace mission planning or human authorization; it supplies the containment and promotion substrate those functions run on.

6. Disclosure Scope

The technical subject matter attributed to the Forecasting Engine in this article, the planning graph as a first-class structure, the containment layer and delusion boundary with immutable speculative markers, the branch classification taxonomy (eligible, introspective, delegable, pruned) and lifecycle states, the governed promotion gate, and executive-graph aggregation, is disclosed in United States Patent Application 19/647,395. This article is a dated public description of that disclosure and its application to governed mission planning.

All statements about Anduril, Lattice, Palantir, and any other named product, company, program, or policy are external context drawn from publicly known information. They describe third-party systems and the surrounding market and are provided for comparison only. They are not claims of the filing, are not asserted as the inventor's own work, and should not be read as characterizing any third party's internal architecture beyond what is publicly and generally known. Where a competitor capability or limitation is described, it is stated at the architectural level; nothing here should be read as an assertion of a specific defect in a third-party product.