Cannot find anything related to Monte Carlo methods in the source code. I thought Spark implemented a conventional 3DGS pipeline with LoD optimizations (And it seems they do the sorting on the CPU using Rust/WebAssembly because of WebGL limitations)