💭 Do I need 1 observed root node samples latent for actual root nodes (no hidden info), then actual root nodes + action sample successor (due to stochasticity) 2 (observed root node itself + action)s sample successors (one sampler handling hidden info and stochasticity 3 observed root node samples latent for actual root nodes with a 'fixed random seed' (sampling hidden info and stochasticity) and actions from that point have fixed outcomes Would they give the same result?