The multi-physics engine that powers Live Parts solves within the GPU in a non-deterministic fashion (slightly differently from run-to-run). This means that cell operations (i.e. addition and removal) may happen slightly differently during each run of Live Parts, even for the same model, resulting in slightly different designs during reach run. However, running Live Parts on the same model enough times reveals that all the resulting objects are largely similar and that even though two design may take different pathways towards their results, they design generally reach the same overall design pattern.
Check out this article to learn more about the Live Parts process.