The Caps don’t have a whipping boy. They’re on pace for a Presidents trophy and killing it. Jensen has been fine. Everyone has been fine.
These types of stats are not reliable indicators. X goals is trash because it’s based on shot distance but not quality. It is by definition dumb, and people are wildly misusing them.
There is more to expected goals models than just shot distance and location.
RPubs - A New Expected Goal Model for Predicting Goals in the NHL
Shot distance, angle, type, duration since the last shot, number of skaters, score state, whether it's a rush attempt, etc. are all contributing factors. I agree that expected goals models aren't sufficient to capture the true value of every shot but it's a decent start and a relatively good predictor of future team success, certainly a better predictor than past team success.
I think the bigger problem with just using on-ice xGF isn't the xGF model itself but rather how to factor in individual skaters' contributions to their on-ice stats given teammates, opposition, score-state, zone starts, etc. Carlson's xGF numbers aren't great, but is it because he himself is influencing the xGF numbers negatively or is it because of his context (teammates, opposition, deployment, coaching, etc.)?
A great example is the play of Orlov and Niskanen this season compared to last: both seem to be doing well this season on different teams and in different contexts but last year together they were a disaster (but before that they were fantastic together!). What caused these changes? It certainly wasn't a poor xGF model, something about them was fundamentally broken while together and everyone here saw this last year.
Carlson's on-ice defensive numbers are quite poor and from my eye-test the team isn't very good defensively when he's on the ice, but like others I don't really think Carlson himself is responsible for his poor on-ice numbers (again, from my eye-test he's seemed good in his own zone). Separating individuals' contributions is significantly harder and this is where I think the next step in analysis is heading. People are trying it through various WAR/GAR models but they still seem insufficient.