First of all "blunder" does not necessarily mean, that the move is obviously wrong. There are obvious "inaccuracies" and tricky "blunders". Also a better engine eval does not necessarily mean, that the move is "better" in a practical game: the "optimal" line might be more difficult for a human (in the way, that you have to find some tricky non-intuitive moves)
@drwerewolf said in #1:
and a few moves later when I went from +14.6 down to +6.3. Those feel pretty BLUNDERific to me
The position is interesting. I certainly would not have played your "+6.3 move" Nxd5, because after a few seconds I still don't see a trivial win after Bxd5 Qxd5 Qxe3+ and Qxd3. The machine recommendation h4 looks ofc good. Personally I would have played quickly the greedy and simple hxg4 instead and the machine likes that move too (+12 eval).
@drwerewolf said in #1:
and a few moves later when I went from +14.6 down to +6.3. Those feel pretty BLUNDERific to me
The position is interesting. I certainly would not have played your "+6.3 move" Nxd5, because after a few seconds I still don't see a trivial win after Bxd5 Qxd5 Qxe3+ and Qxd3. The machine recommendation h4 looks ofc good. Personally I would have played quickly the greedy and simple hxg4 instead and the machine likes that move too (+12 eval).