Yes, the computer evaluations are extremely useful. But they are measuring the quality of the position when two computers are playing. Yet most of the games you care about tend to be two humans playing each other. And those humans do not play like computers. The computer might say the game is even, and maybe it is with perfect play, but one side can be much harder (easier) to play than the other. So I suggest this trick. Go to analysis.sesse.net, which covers top games (only). Scan down the vertical list of all possible moves and consider the distribution of outcomes. If the top move is great for White, but all the others are not, robustness is low, especially if the top move for White is not super-obvious (such as recapturing a Queen, etc.). If all the sequences look very good for White (Black), you will know that for humans the position probably is somewhat better for White than the single computer evaluation number will indicate. Robustness against human error will be present.
For the Carlsen match, here is a good Twitch stream, currently with Caruana as commentator.