Why there is no rigid-body solver here
A physics engine is the obvious way to build a domino run and the wrong one. A general solver steps on wall-clock time and its contact resolution is sensitive to iteration order, which means the live preview and the offline render will diverge.
For most modes a small divergence is invisible. For a chain reaction it is fatal, because the whole video is a single causal sequence: if one contact resolves differently, everything after it is different, and the export is not the video that was previewed.
So the cascade is computed deterministically rather than solved. The trajectory of each topple follows from the layout and the spacing rule, which is why the run can be trusted to reproduce exactly and why the trigger can be placed on a beat with confidence.

