PosteriorBench targets scientific inverse problems, where indirect or partial measurements can have multiple physical explanations. Its researchers say they built high-fidelity reference posterior distributions—describing possible solutions and their probabilities—across four tasks: Darcy flow inversion, Poisson source recovery, carbon capture and storage, and light transport material inference. Five complementary metrics let researchers evaluate their solvers against those references. The team reports that better reconstruction accuracy can coincide with worse posterior recovery, with even strong generative samplers struggling to capture both the mean and variance. They also shared links to the paper and code.