I've had to look back at some recent games because I've lost some in ways I didn't figure out right away or that were just particularly stupid.
#1 -- sacrifice ambush
I had pretty much shut my opponents bishops off on move 5 but I then had to put quite a bit of resources on keeping safe the base pawn on F5 and ended up being quite passive altogether. What ended up killing me was the idea of a) getting my queen moving through the big diagonal that it did share with the opponent's bishop but which I nevertheless thought was safe because of that little barrier of pawns way up in my opponent's territory -- as well as b) getting castled in that same corner targeted by the white bishops, which, again, I thought would still be safe.
No, dude. 16.Nxe4. All downhill from there.
How should I have taken such a sacrifice into account? The clue, I suppose, is that exactly because I had put white's bishops into such a nasty position, I should expect my opponent to put something a bit more elaborate than the average planning into how to get them free. An additional data point: Here's how the Stockfish engine played the game from move 11 onwards instead of going with the queen development idea I used. Note that white still makes the same sacrifice even though it can't use it to threaten the queen. Now black wins.
(Even though chesspastebin, AFAIK, doesn't really support chess960, I was able to display this one because it happened to not have any castling.)