I have written before about the debug advice to “Quit thinking and look.” It means that you should not form conclusions prematurely. Stop and look at what is going on instead of guessing and cooking up theoretical scenarios. Sound advice that I completely failed to follow in the case that I just chronicled on my Intel Blog: https://software.intel.com/en-us/blogs/2020/03/18/quit-thinking-and-look-chasing-simics-performance
I suspected temporal decoupling, while the answer turned out to be about stalls added to certain instructions to optimize polling loops. In this case, with software that was not polling on the instructions, it was a pessimization. By changing the Simics settings, the execution was sped up many many times.