The validation process in X-Cube-AI consist on running the model using the original framework and comparing this run with a run of the C generated code.
When you use random data the result may be good or bad but it doesn't tell you a lot of the actual results you are going to have with the real data (the random data may be far from your actual ones). It is a quick way to see if the C code generation is behaving correctly.
You can refer to this chapter of the documentation Documentation/evaluation_metrics.html#ref_l2r for a detailed L2R formula.
In STM32CubeMX you can open the documentation using the Help->X-Cube-AI Documentation menu
If you want to access it directly it is in the pack distribution. Usually on Windows it is under
C:\Users\***\STM32Cube\Repository\Pack\STMicroelectronics\X-Cube-AI\5.0.0\Documentation
Hope it helps
Regards
Daniel