ZEngine — Behavior Trees

Priority: Next-year plan Status: Design — no behavior-tree runtime, asset schema, ECS component, or visual authoring tool is implemented. Depends on: game-logic extension boundary, animation/physics contracts where used, and durable scene/asset serialization.

Direction

Behavior trees are one possible AI authoring model. The old node layout, blackboard, callback table, scheduler registrations, Lua comparison, and performance figures were proposals—not existing engine APIs or measurements.

Required design

• Define whether trees are authored assets, code-defined graphs, or both; establish stable schema/version/migration and reference rules.
• Specify node semantics, evaluation budget, deterministic/random behavior, tick rate, cancellation/abort rules, debug tracing, and failure handling.
• Keep per-agent mutable state separate from shared definition data and give both clear scene/asset/reload lifetime rules.
• Expose gameplay/physics/animation interaction through stable interfaces, not direct assumptions about unimplemented subsystem types.
• Define a blackboard type system and bounded memory policy before publishing a designer-facing format.

Acceptance gates

• Deterministic tests cover composites, decorators, aborts, long-running actions, invalid graphs, budget exhaustion, and scene/asset reload.
• Visual debugging and profiling demonstrate the documented update cost and lifecycle behavior for the target game workload.