Execution and caching

How a graph becomes a run, and why the second run is faster.

Ordering

The engine sorts by execution phase, then topologically within each phase. A cycle is refused with SS-E011 rather than hanging.

Caching

Each card's output is cached against a key built from its inputs and its resolved parameters. Change one parameter and only that card and its dependants re-run. Change nothing and nothing runs.

This is why a long time-frequency step costs you once rather than on every adjustment downstream of it.

Progress and cancelling

Cards report progress while they run, and Escape stops the current run. Partial results already cached are kept.

When it runs out of memory

The usual culprit is time-frequency output, which is trials by channels by frequencies by samples. SS-E060 explains what to reduce first, and the recipe works through it.