In three online experiments involving 300 US adults, researchers varied how reliably rewards matched their labels, how often rewards changed, and whether chosen movements went as intended. Participants responded faster before their first choice as randomness increased, while computational models suggested they used simpler, less precise decision policies.