The aggregates are more meaningful, but if we persistently conceded goals from lower xG situations versus higher xG situations, over time it would suggest the concept was misrepresenting those situations.
xG is presumably set up so that aggregate expected goals should approach aggregate goals over time given the body of prior data, otherwise it'd be a bit weird. Perhaps Claude or Quincy can confirm that.
Mbeumo's near post strike was both clever and well executed, so it's a goal. I'd be interested to see the post shot / goal frame analysis on it.