This is however a problematic model to sample. Your priors should try to exclude impossible values, so the sampler doesn’t find them, or you should reparametrize so the data is on a more unconstrained space (like how positive data is usually modeled on a log-scale, or equivalently using an exponential link)