Build the matchup. Then try to break it.

Start with both divisions

Build the attacker and defender, then set the battlefield assumptions you care about. Armor and piercing, hardness, soft and hard attack, defense or breakthrough, organization, width, terrain and equipment cost can all change the result.

Use Counter for a specific problem

If your division is losing, Counter tests structural changes against that exact opponent. Normal search stays bounded to one or two changes. Optional deep redesign can try a third change when the smaller search cannot find a meaningful answer.

Read the recommendations

The recommendation cards separate raw combat performance from efficiency and from the smallest change that clears the improvement threshold. The site also shows the current and proposed equipment cost as X IC/div → Y IC/div, plus the absolute and percentage change.

That cost is context, not a production gate. Counter does not inspect your saved factory lines or decide whether your current economy can afford the template.

What counts as meaningful?

A recommendation needs at least a +2 percentage-point modeled win-rate improvement. Final recommendations are battle-simulated rather than chosen from a simple stat heuristic, and strength-loss exchange can break close ties where useful.

Accuracy boundary

Some inputs are game-file exact, some runtime behavior is executable inferred, and the recommendation/search layer is planner analytical. The planner does not claim bit-for-bit parity with hoi4.exe. See the methodology page for the current boundary.