Multi-domain wargaming and course-of-action analysis

A staff developing courses of action can usually war-game two or three branches properly within the time available. The adversary is not constrained to those branches. Manual wargaming is also uneven: the quality of the red picture depends on who is playing red that week, and the assumptions that drive the outcome are often implicit, recorded in nobody's notebook, and impossible to revisit when the situation changes.

What planners need is not a machine that picks the plan. It is the ability to test many more branches than a staff can, to see which assumptions the outcome is sensitive to, and to have that reasoning written down in a form that survives a change of shift.

The wargaming profile draws on planning data the headquarters already maintains:

The Red-Force Behaviour agent generates adversary responses from doctrine and observed practice rather than from a single scripted enemy plan. The Blue-Force Posture agent maintains the friendly picture, including what degrades when sustainment or communications are contested. Scenario Synthesis expands each candidate course of action into a branch-and-sequel tree.

The COA Analysis Engine then evaluates that tree using a combination of Monte Carlo sampling, Bayesian updating on uncertain parameters and reinforcement-learning agents for adversary play, producing probability-weighted outcomes rather than a single predicted future. A sensitivity pass identifies which input assumptions, if wrong, change the ranking, and the doctrine-alignment function of SC-STR-01 Strategy & Doctrine Agent checks each option against applicable doctrine and constraints.

Planners set the problem in their own terms: the mission, the constraints, the assumptions they are prepared to make. They can state these in natural language and see them reflected back as an explicit, editable assumption set, which matters because the assumption set is what the analysis is actually about.

Results are explored, not merely read. A planner can pin a branch and ask what happens if the adversary commits reserves twelve hours earlier, if the port is unavailable, if satellite communications are denied for six hours. Each answer names the assumptions it rests on and the confidence attached to them.

The platform does not select a course of action, and it does not present a single recommended plan without alternatives. It ranks options against criteria the staff set, shows why the ranking came out that way, and makes it easy to disagree with it.

The commander's decision is recorded as a decision, with its supporting analysis version, so that a later review can distinguish between a plan that failed and a decision that was unreasonable on the information available at the time.

A wargaming run leaves the staff with:

Order of battle and logistics state are ingested from existing planning and C2 systems; products are exported in formats the headquarters already staffs, including briefing-ready comparison matrices and structured data for the operations plan. Where a headquarters runs a Maven-class or national C2 environment, SynapseCommand is positioned as an analytical layer beside it rather than a replacement for it.

Wargaming is compute-hungry, so this profile is usually hosted in a national sovereign cloud or a headquarters data centre with GPU capacity, inside the relevant accreditation boundary. A reduced-fidelity configuration runs on deployable hardware for use in the field, trading the number of sampled branches for footprint.

The practical gain is coverage. A staff that could examine a handful of branches can examine the space around them, and can re-run the analysis when an assumption changes rather than starting again. Because assumptions and sensitivities are explicit, review and handover are faster and less dependent on individual memory.

Figures quoted anywhere on this site are illustrative planning targets from internal test scenarios, not certified operational results.

A wargame is only as good as its model of the adversary and its input data. Doctrinal red play captures how an adversary is expected to fight, not how a specific commander will choose to fight on the day, and genuinely novel behaviour is by definition under-represented.

Probability-weighted outcomes are conditional on the stated assumptions; they are not forecasts. The platform is deliberate about saying so, and shows the assumption set alongside every result rather than presenting bare numbers.