Tracked the accuracy of my F-Series Rolex Sub for about 30 days a couple of years ago, and it was running around -1.0 second/day. Same movement, but prior to the current Submariners being COSC certified. More recently (but less formally tracked), I've found it gains about 1.0 second/day.
Remember also that COSC certifications, like cheese, only speak to validity for a certain period of time. Some say that all it means is that the mechanism managed to pass the test when it passed the test.
You'll also find that performance can vary based on a number of external factors. My Sub ran differently on its OEM bracelet, versus NATO strap (where the caseback was separated from the heat of my wrist by that slight amount and not influenced heat transfer from the SS bracelet).
Et cetera, et cetera, et cetera.
