Simulation and tuning / lesson 05
Ask whether quality earns its cost
Useful flow compares configurations; sensitivity points to the next knob.
Know what the primitive owns.
Problem
More effort can improve quality while still making the system worse per unit of cost. A score without a local change signal also leaves tuning direction unclear.
System boundary
Callers define quality, cost, weights, and knobs. CMG supplies usefulFlowScore and numerical sensitivity utilities.
State changes become visible.
Inspectable: The chosen knob values, recorded cost slices, bounded quality, useful-flow score, and ranked local gradients.
Sweep effort while parallelism and verification depth stay fixed.
Static specimen / JavaScript adds controls without changing the lesson.
Same idea, honest surfaces.
const { score } = usefulFlowScore(quality, cost);
const ranked = rankSensitivity(
current,
phiOf,
);
console.log(score, ranked[0]?.name);score = useful_flow_score(quality, cost).score
ranked = rank_sensitivity(
current,
phi_of,
)
print(score, ranked[0].name)What this does not imply.
Φ is not an objective truth. It is only as honest as the caller-defined quality and cost terms supplied to it.