Would be interesting to see how those numbers shake out on a with and without you split. Is one better than the other when they're apart? Does that matter?
Short version:
They are RAPM adjusted so it (mostly) accounts for the mins they play together and apart. Think of it as a super WOWY that also includes competition and other factors.
Long version:
RAPM is a regression model used to solve all sets of shifts with 10 unique players (plus other variables added) and the results that occurred over that shift.
So you have one shift of 5v5 and the home team is +2 xG and then you have one player on home team switch for a new player and the team goes +1 over that time frame, you'd get two shifts:
aX1 + bX2 + cX3 + dX4 + eX5 - zY1 - yY2 - xY3 - wY4 - vY5 = +2
aX1 + bX2 + cX3 + dX4 + fX6 - zY1 - yY2 - xY3 - wY4 - vY5 = +1
What RAPM does is take every single shift of unique 10 man units like the two you have above, and then solves for these factors. Adjusted Plus Minus (RAPM's god father) just solved using regular least squares mean. Regularized Adjusted Plus Minus merely uses ridge regression as opposed to least squares mean.
Ridge regression is a Bayesian filter which makes up for some issues with LSM.