That may be the case, but simply running the same code review twice with the same model may give very different results as well. Impossible to know with the information we have.
today we can use some anecdata based on our observations, this also helps shape the dimensions we want to eval the agents-tasks on