Run one generation repeatedly¶

This page continues from Generate, build, and run. It assumes the generation at $GENERATION_DIR already contains generated/ and build/main.

Why generation and run IDs differ¶

A generation identifies one authored source snapshot and its derived RDF, IR, C++, contracts, provenance, and build. A run identifies one execution of that build. Changing the model requires a new generation; repeating an experiment with the same build requires only a new run ID.

digraph reuse {
rankdir=LR;
graph [bgcolor="white", nodesep=0.35];
node [shape=box, style="rounded", fontname="sans-serif"];

generation [label="generation\ngenerated/ + build/"];
run1 [label="runs/run-1\nlogs + REC + runtime RDF"];
run2 [label="runs/run-2\nlogs + REC + runtime RDF"];
run3 [label="runs/run-3\nlogs + REC + runtime RDF"];

generation -> run1;
generation -> run2;
generation -> run3;
} — Figure 5 One immutable generation owns multiple runtime records.¶

Create another run¶

$ motion-spec run "$GENERATION_DIR" --run-id run-2 --headless

Only runs/run-2 is new. Its manifest references the existing generation artifacts relatively; the source, model, controller, contracts, provenance, and executable are not copied into every run.

Bound a run¶

$ motion-spec run "$GENERATION_DIR" --run-id run-3 --headless --steps 20000

Everything after -- belongs to the executable, not to the CLI.

Each invocation creates a fresh run directory. This preserves each REC lifecycle and prevents accidental log replacement.

When to generate again¶

Create a new generation when the .robmot, scene, FSM, code-generation inputs, or target configuration changes. Do not create a new generation merely to repeat an unchanged controller execution.