ProCredit Enables Verifiable Progress Rewards in Agentic RL

ProCredit moves beyond sparse outcome rewards by using intermediate state acceptance checks to provide training signals for long-horizon tasks. This method allows agents to receive credit for verifiable progress during tool-use sequences rather than waiting for final task completion.