Proud to share two preprints! muCode presents a simple and scalable method for multi-turn code generation leveraging learned verifiers. ORCA formulates rewards as an ordered coverage problem, enabling robots to learn from a single temporally misaligned video demonstration.