How this is worked out
The part is split into small cubes (voxels) and solved as a linear-elastic body, using conjugate gradients with a multigrid preconditioner. Walls and top/bottom layers count as solid plastic; infill is a softer core whose stiffness scales with density squared. The walls and shells are checked against the filament's strength along the layers, and against layer adhesion where stress pulls the layers apart. Near sharp corners, stress is averaged over a 1 mm ball before it is compared with the filament strength (the critical-distance method; 0.93 mm was measured for printed PLA), so tiny stress spikes the grid size creates don't decide the result. Points touching a fixed support are left out of the check, because a perfectly rigid support is an idealisation.
Treat it as an estimate. Real prints vary with temperature, cooling, moisture and print quality, and loads that stay on can make plastic creep. It doesn't model impacts, fatigue or heat. Don't rely on it for anything that could hurt someone if it failed.