Dense Progress Rewards for Long-Horizon Agents: How Tev Turns Multi-Day Tasks Into Trainable Checkpoints

How reward shaping and intermediate checkpointing enable agents to learn economically valuable tasks spanning hours to days.