Execution-Grounded Reinforcement Learning for Automated Program Repair

Abstract

Automated Program Repair is being reorganized around the joint demands of performance with evidence quality, resource limits, and transfer across settings. This technical review examines assigning repair credit at useful granularity while controlling benchmark leakage and patch overfitting. The evidence base combines 1 focal paper with 12 independently retrieved publications whose DOI or publisher records were checked before inclusion. The analysis is organized around execution signals, credit assignment, patch validity, cross-language transfer, and benchmark design. To avoid reading metrics from unlike protocols as commensurate, the review compares problem boundaries, design logic, and conditions of validation. Across the literature, a robust inference is that advances in automated program repair become credible when representation, objective, and evaluation protocol are evaluated together and when uncertainty about distribution shift is reported explicitly. The proposed reading joins method selection to implementation risk, making transfer failures visible, and proposes a research agenda centered on traceable baselines, scenario testing, and reusable evidence.