What is the difference between a brain and a doctrine?
A fight takes place across three lanes, with towers protecting each side’s core. The commander chooses when to hold and when to spend energy spawning a unit into a lane. Its goal in ranked competition is to destroy the enemy core while protecting its own.
The brain is the structure making those decisions. Doctrine describes preferences such as pressure, defense and trades. Training is the process that changes a commander. Those are different concepts, and none requires you to write scripts.
How training makes those changes is being reworked. The relationship between scoring preferences and training updates depends on the implementation. Use the options and explanations in the current workspace, then check what the recorded fights actually show.
| Choice or evidence | What it tells you |
|---|---|
| Brain | How the commander represents decisions. |
| Doctrine | Which behavior the scoring preferences describe. |
| Training results | What a recorded candidate did in the selected fights. |
| Ranked results | How the published version performed in competition. |
Dense brain
Dense is the default brain family. Layers of connected values turn the current fight information into an action. Its layer settings change the size of that decision-making structure.
If you are unsure where to start, the default gives you a first commander to inspect. A larger brain has more parameters to train, but size alone does not establish that it will win more fights.
Lane specialist brain
Lane specialist processes information for each lane before combining it to make a decision. Its structure follows the three-lane battlefield more directly than a Dense brain.
The name describes how it processes information. It does not mean it must attack one lane or that it has an advantage against every other brain family.
Memory brain
Memory carries short-term state within a fight. Its decisions can use information from earlier moments as well as the current battlefield.
That memory belongs to the current fight. It does not mean a published commander keeps learning from one ranked fight to the next. Training and publishing an updated version are separate actions.
Rulebook brain
Rulebook uses trainable thresholds and decision rules rather than a layered neural structure. Training adjusts those values to search for useful behavior.
Rulebook is still a way to build a commander without coding. You use the game interface instead of writing the decision rules as your own program.
How should I compare brains and training doctrines?
Start with a question and keep the comparison focused. If you change the brain, scoring preferences and training duration together, a different result will not tell you which change mattered.
Use Analytics to examine the selected run and generation, then compare compatible versions against the same opponents in Tests when available. Check coverage and fight counts. Raw fitness from different scoring objectives is not a shared measure of strength.
There is no universal best brain established by its name or parameter count. Ranked outcomes measure wins, draws and losses, while the other measurements help you understand the fights. Choose the next experiment using that evidence.