Anduril's Lattice Plans Missions Without Speculative Containment

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 deployments, on Royal Navy and Royal Australian Air Force 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. An exploratory branch that imagines a flanking maneuver must be structurally distinguished from a viable branch that has been partially validated against current intelligence and from a promoted branch that has cleared the activation threshold. All three exist in the planning graph. They carry different promotion thresholds, different operational implications, and different operator-facing semantics. Without classification, all speculative branches look the same to the system, and the operator cannot distinguish mature alternatives from raw speculation, which collapses the cognitive value of having alternatives at all.

Anduril cannot patch this from inside the current Lattice architecture because forecasting in this sense is a planning substrate, not a planner feature. Adding more course-of-action generators, better optimizers, or richer scenario libraries does not produce contained speculative branches with classification and promotion semantics; it produces a higher-resolution version of the same generate-and-select pipeline the platform already runs. The forecasting-engine shape is architectural, and the Lattice planning layer's shape is fundamentally that of an optimization engine over a heterogeneous asset graph.

3. What the AQ Forecasting-Engine Primitive Provides

The Adaptive Query forecasting-engine primitive specifies that every governed planning system maintain a planning graph as a first-class cognitive structure, with speculative branches under structural containment, branch classification by maturity and risk class, governed promotion thresholds, and an executive-graph aggregation property that records the cognitive history of planning across the mission. Planning is not 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. A speculative branch is a region of the planning graph that has read access to the active situational state but no write access to the active plan or to the assets executing it. Branch evaluation runs against a forked projection of the world; the projection evolves as the active world evolves; the projection cannot affect the active world. This isolation is not a coding convention; it is a structural property of the substrate, enforced by the same mechanism that enforces credentialed observation in the broader AQ chain.

Branch classification is governed by a defined taxonomy: exploratory branches imagine options without intelligence backing; analyzed branches have been evaluated against current intelligence and asset state; viable branches have cleared feasibility thresholds; staged branches have cleared promotion thresholds and are positioned to activate on a defined trigger; active branches are the executing plan. Each class carries different promotion thresholds, different operator-facing semantics, and different interaction rules with intelligence updates. Reclassification is automatic when conditions change — a viable branch whose key intelligence assumption is invalidated reverts to analyzed or is pruned — and every reclassification is recorded as a credentialed observation in the executive graph.

Promotion is the load-bearing structural mechanism. A branch can advance only by clearing its class-specific threshold, which composes feasibility, intelligence-confidence, rules-of-engagement compliance, asset-availability, and authority-credentialed approval where required. Promotion to active requires the human-on-the-loop signature that defense doctrine demands; the substrate makes that signature a structural gate, not a UI checkbox. The executive-graph aggregation records the full cognitive history — why was this branch promoted, what intelligence triggered the reclassification of that alternative, when did the system first consider the option that ultimately became the active plan — and is essential for after-action review and for building institutional planning knowledge.

The primitive is technology-neutral with respect to the underlying optimizer (classical, learned, or hybrid) and composes hierarchically — squad-level planning graphs roll up into platoon-level and theater-level graphs under the same shape. The inventive step disclosed under USPTO provisional 64/049,409 is the closed forecasting-engine substrate with contained branches, classified maturity, governed promotion, and executive aggregation as a structural condition for safety-governed autonomous planning systems.

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 — U.S. services, Five Eyes partners, NATO members, Indo-Pacific allies — who need auditable evidence that the autonomous-planning systems coordinating their assets maintained governed speculative containment at every moment. The substrate belongs to the customer's authority taxonomy and is portable across hardware refreshes and software upgrades, which paradoxically makes Lattice stickier because the platform's fusion and effector value is what differentiates its access to 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 the long-running concern, voiced both by the Defense Innovation Unit and by allied procurement bodies, that autonomous-planning systems remain optimizers rather than governed planners; a defensible position against in-market competition from Palantir's MetaConstellation and Maven Smart System and from incumbent primes' homegrown stacks by elevating the architectural floor; and a forward-compatible posture against emerging DoD policy on autonomy-test-and-evaluation, the 3000.09 directive on autonomy in weapons systems, and allied frameworks that are converging on evidence of governed speculation rather than mere human-on-the-loop UI. What customers gain: auditable planning-graph lineage, portable across mission and theater, expressible in a single primitive shape from squad to coalition. Honest framing — the AQ primitive does not replace mission planning or human authorization; it gives both the substrate they have always needed and never had.