NeuroStacks

Composition model

Stacks are graphs, not free-form scripts. Nodes are typed pieces (sensor, adapter, neuroblock, control, state, actuator). Edges connect ports. The graph checker rejects cycles where forbidden, writer conflicts, missing required inputs, and type mismatches.

  Sensor ──▶ NeuroBlock ──▶ Control ──▶ Actuator
                │
                └──▶ (optional) State / Adapter

Identity

Stack identity is cryptographic over structure and locked modules — not over canvas layout. Moving nodes on the Compose board does not change the digest. Changing a module version or an edge does.

Executable vs source packages

Some exports are fully runnable under the current composed runtime. Others are source or planning packages marked executable:false. Always read the manifest: SynapseVM will not pretend a planning graph is a finished on-robot controller.