Neural Code Translation of Legacy Code: APL to C#

This collection of recent research covers LLM-enabled software engineering and testing, with practical implications for robustness, safety, and integration risk. Papers present advances in neural code translation (APL→C#), model-merge editing (CRANE) that alters reasoning/tool-use tradeoffs, code-aware agents for targeted game testing (CA2), and LLM-driven strategies for robustness testing and metamorphic testing of LLMs and microservices. Empirical studies highlight that prompt strategy (not model size) often drives test diversity and failure-mode coverage, guided few-shot prompts performbest