Examples

Runnable examples that train real models on canvas-structured data. Each example generates visualizations and demonstrates a specific capability.

Core examples (v1 — compile_schema)

#	Example	What it demonstrates	Key feature
01	Hello Canvas Types	Declare, compile, train, visualize	Field, compile_schema
02	Multi-Frequency Fusion	Structured vs flat allocation comparison	Bandwidth-proportional allocation
03	CartPole Control	Real gym environment, BC + consistency loss	Self-consistency feedback dynamics
04	Vehicle Fleet	64-vehicle cooperative trajectory prediction	Multi-agent with social forces
05	Protein Complex	4-chain binding affinity prediction	Molecular structure as canvas types
06	Air Traffic Control	Conflict detection with 12 aircraft	Safety-critical multi-agent
07	Hospital ICU	Full-ward simulation with 6 patients	Deep type hierarchy
08	Minecraft World Model	Next-frame prediction with imagination	Temporal hierarchy (period=1/4/16)
09	Brain-Computer Interface	Multi-modal neural decoding (M1/PMd/S1)	Multi-array cursor + speech decoder
09b	BCI + TRIBE v2	Canvas decoder on real cortical predictions	Modal GPU, TRIBE v2, 68.8% accuracy
10	Fusion Reactor	Disruption prediction + plasma control	Multi-timescale diagnostics
11	Mars Colony	Cascading failure detection	5+ subsystems, alert prediction

v2 examples (compile_program — families, carriers, operators, scheduling)

#	Example	What it demonstrates	Key v2 feature
14	Saccading Vision	Event-triggered scheduling, residual-driven fovea	RegionScheduler, ResidualAccumulator, families
15	World Model Carriers	Mixed deterministic/diffusive/filter dynamics	carrier field, compile_program
16	Memory Consolidation	Boundary clocks, working→episodic→semantic	ClockSpec, ProgramCompiler, compile modes
17	Graph Parser	Operator types, fixpoint iteration	operator, ClockExpr IR
18	Sparse Reflection	Uncertainty-driven self-reflection	ConstraintSpec, LearnedScheduler

Research examples (real experiments with published results)

#	Example	What it demonstrates	Key result
19	Cortical Dynamics	Brain dynamics prediction with TRIBE v2	R²=0.825, 23 regions, 42 cortical pathways

Running

# Install canvas-engineering
pip install canvas-engineering

# Run any example (CPU, 30-120 seconds)
python examples/01_hello_canvas_types.py
python examples/14_saccading_vision.py

# Run TRIBE v2 BCI on Modal (requires GPU)
modal run examples/09b_bci_tribe_modal.py

# Run all examples on Modal
modal run run_examples_modal.py

Each example generates a multi-panel visualization to assets/examples/.