Coalgebras of measurable spaces are of interest in probability theory as a formalization of Labelled Markov Processes (LMPs). We discuss some general facts related to the notions of bisimulation and cocongruence on these systems, providing a faithful characterization of bisimulation on LMPs on generic measurable
spaces. This has been used to prove that bisimilarity on single LMPs is an equivalence, without assuming the state space to be analytic. As the second main contribution, we introduce the first specification rule format to define well-behaved composition operators for LMPs. This allows one to define process description languages on LMPs which are always guaranteed to have a fully-abstract semantics.
