HABIT-Bench Evaluates Habit Inference in Long-Term Agent Memory

HABIT-Bench measures an agent's ability to infer implicit user habits from weak, longitudinal evidence across hundreds of dialogue sessions. The benchmark tests whether memory systems can distinguish between stable user preferences and temporary exceptions in tasks like finance and software.