Trying to figure out what a blunder is in my data

[regal sigil]

Is there another variable used by sf that determines blunders that I'm not taking into account?
Stockfish doesn't determine what a blunder is, that is done by the interface (ie Lichess, Chess.com, etc).

I've seen definitions of a loss of 30 or more in Win%
That is what Lichess uses, yes

in my dataset some blunders are as low as a loss of 2.25 in Win%
That shouldn't be the case, but maybe your dataset has some subjective blunders. For example, playing Na3 on move 1 is technically not a game-losing blunder.

[uneven geode]

I think you may be right about the final point about subjective blunders. I've just run an equation which was Change in Win%/original win% and the lowest proportional change was a 40% decrease in Win% so that may be the definition the software I've used has done. Thank you for the help : )

Okay I've found out that the minimum proportional decrease in Win% for blunders is 42.92%, and the maximum decrease for non blunders is 42.85%. Does anyone know if 42.9% is a significant measure for any known interface? Thanks!

[regal sigil]

Doubt it... maybe you are measuring it wrong