Worked on the volcengine/verl repository to address a bug affecting Prime reward calculation accuracy. Focused on backend development using Python, the work involved enhancing asynchronous compute functions by adding support for extra_info in both single and parallel score computation paths. This ensured that PrimeRewardManager consistently retrieved and passed extra_info during score calculations, improving the reliability and consistency of Prime mode rewards. The solution leveraged asynchronous programming and distributed systems concepts to maintain correctness across different workflows. All changes were linked to a specific commit for traceability, reflecting a methodical approach to debugging and maintaining robust reward computation logic.