A compact, focused model at 1.5B parameters built around a specific editing task — predicting the next edit in a sequence. It operates on a narrow but practical problem: given prior changes, what comes next? Its small footprint makes it fast and deployable, though its specialization means it's not a general-purpose assistant.