I agree - all attempts at automatic classification of blunder have same problem. This is why analysing games without engine still matters and is going to matter for foreseable future.

Don't forget also impact of time control - shorter games lead to more mutual mistakes. While in 90+30 first big blunder should decide the game, in blitz it's just the beginning.

Amusing example is Chessbrah speedruning to 2000, while hanging queen in every game: https://www.twitch.tv/videos/593176969