Crowdsourcing and human computation integrate cognition into computationally mediated workflows. Here, we introduce Judicious, a suite of human-in-the-loop elementary random procedures that can be inserted into probabilistic programs to run and model closed-loop crowdsourcing and human-computation experiments. Each procedure abstracts the procurement of a human decision into a single blocking function call, thereby enabling flexible and idiomatic commingling of inference and behavioral data collection. We discuss the architecture of our Python-based implementation of Judicious and provide examples of tasks and algorithmic workflows.