更少的token，无需执行MAS得到效果可比的RM和workflow；
即： agentic workflow Evaluation;
basic setting:
Optimizer as Reward Model； or Optimizer-Based Reward Model;
❌Query-Level； ✅Dataset-Level （效率更高→几次重复取avg)
Static Code Analysis;
Automated evaluation of code generation tasks;
Evaluating Code Intelligence Tasks;