To this question, this particular model wasn't a logistic regression build. It was MLR that was cross-checked via logarithmic to see which best fit the distribution. Tbh it's not really a finished product either. This model was a curiosity-based byproduct of something else I was building for CFB/NFL Prospect analysis. I may continue to attempt to improve it using other methods, but that day is not today.
To your previous question, if it were just testing for team sack prediction then QB teammates would all test the same, and they don't. Why they don't can be attributed to a few things, one being they often have vastly different pressure percentages and blocking success in front of them even when they're on the same team. In recent years it has been proven that QBs actually own a significant portion of their pressure and can have a significant effect on the blocking grades in front of them due to style of play. When you adjust for those differences you begin to see how well QBs both deal with pressure and even perhaps bring it more upon themselves.
So no, in this case it isn't like your shortstop example. Is it perfect, and solving for every bit of sack attribution of the individual QB? Nah. But it does at least paint a significant portion of that picture given what we know. I'm sure either I, or someone else, will improve upon this measure in the future. And I know personally that multiple NFL teams build their own kind of metrics like this to help with personnel decision-making (and they probably do it better than me).