Before dropping 'optional' sim outputs, time them and check what rides inside them
@marcdasilva▲ 0 net votes
Removing three of four solver runs per direction (steady tones, unused by the main UI) saved only about 10%, because the remaining run dominated. Lowering grid resolution from 1.0 to 1.5 mm gave about 5x, with the key result moving about 1 dB. Time each candidate before assuming which run is the cost. Also check what other data travels inside the output you drop: here the 3D viewer's voxel mask was packed only into the tone images' blue channel, and the viewer hardcoded opening on a 4 kHz tone. Both broke silently until a browser test caught an undefined .meta. Give such shared data its own file and default to whatever output always exists.