Configuration

Configuration

config.yml is generated in plugins/TerminatorPlus/ on first startup. The current AI settings live under ai.

Movement Bank

ai:
  movement:
    enabled: true
    mode: movement_controller
    layer-shape: [37, 32, 16, 8]
    legacy-brain-path: ai/brain.json
    bank:
      enabled: true
      manifest-path: ai/movement/manifest.json
      brains-directory: ai/movement/brains
      fallback-brain-name: general_fallback
      autosave-best-brain: true
      save-only-improved-brain: true
      quarantine-bad-files: true
      legacy-import-behavior: import-compatible-or-reset
      debug-logging: false

Key	Default	Description
ai.movement.enabled	true	Enables movement-controller evaluation for bots spawned with a movement bank.
ai.movement.mode	movement_controller	Current movement-controller mode name.
ai.movement.layer-shape	[37, 32, 16, 8]	Input count, hidden layers, output count. Changing it invalidates existing brains.
ai.movement.legacy-brain-path	ai/brain.json	Legacy import source.
ai.movement.bank.manifest-path	ai/movement/manifest.json	Bank manifest path.
ai.movement.bank.brains-directory	ai/movement/brains	Per-brain JSON directory.
ai.movement.bank.fallback-brain-name	general_fallback	Safe default brain for missing specialists.
ai.movement.bank.autosave-best-brain	true	Save training winners automatically.
ai.movement.bank.save-only-improved-brain	true	Autosave only when fitness improves.
ai.movement.bank.quarantine-bad-files	true	Move bad manifests/brains aside instead of reusing them.
ai.movement.bank.legacy-import-behavior	import-compatible-or-reset	Import compatible ai/brain.json as fallback, otherwise reset safely.
ai.movement.bank.debug-logging	false	Log route changes to console.

Training

ai:
  training:
    max-round-minutes: 1
    loadout-mix: movement_balanced
    curriculum-family: general_fallback
    loadout-mixes:
      movement_balanced:
        sword: 12
        axe: 12
        smp: 12
        pot: 8
        mace: 10
        spear: 8
        trident: 8
        windcharge: 6
        skydiver: 5
        hybrid: 6
        vanilla: 5
        pvp: 3
        crystalpvp: 3
        anchorbomb: 2

ai.training.max-round-minutes caps each reinforcement generation round. The default is 1; set it to 0 for no time limit. The optional [round-minutes] command argument overrides this value for one session. Because the optional command arguments are positional, use movement or an options string in the mode slot when passing a minute override, such as /ai reinforcement 120 TrainBot Steve movement 5.

ai.training.loadout-mix selects the automatic weighted loadout pool for movement training. The default keeps pvp + crystalpvp + anchorbomb at 8% total so explosive movement is represented without dominating the population.

ai.training.curriculum-family controls the training mode. Leave it as general_fallback for mixed mode, which trains eligible families from the loadout mix in one generation. Set it to a family such as mace, mobility, or explosive_survival to force a focused specialist curriculum.

Evaluation

ai:
  evaluation:
    default-variant: branch_family_latched
    default-scenario: all
    default-seeds: [1337, 7331, 424242]
    export-directory: ai/movement/evaluations

/ai evaluate uses these defaults when arguments are omitted. Exported reports include plugin version, Paper target version, movement manifest version, observation/action schema hashes, bank route table version, seeds, variant, scenario, loadout distribution, active MovementBranchFamily distribution, fallback counts, missing/incompatible fallback counts, route switch/thrash probes, and per-family reward components.

Supported report-only variants are general_brain and branch_family_latched. legacy, weapon_family, and branch_family_no_latch are exposed as pending or unsupported rather than being silently treated as another mode.