Analysis tool predictions sometimes not based on best play

Maybe this has come up in detail before (hard to find via a search) but in the analysis tool, it seems that sometimes the predicted move sequences include mistakes and blunders. This is either under "game may have continued", or "x move is best because y". How is this determined?

It will be the estimated best line based on the engine depth.

For clarity, this isn't me thinking that a move is a blunder. The tool itself is saying it's a blunder. So how could a marked blunder be what the tool thinks is the best line?

Without the example, it's hard to say, but a lower depth analysis can miss tings as can some complex positions. Engines don't understand every position the same.

For instance, say this game: https://www.chess.com/analysis/game/live/15447785941. On the Analysis tab, under 'Game may have continued...', if you click 22. Ke5, it shows as a blunder.

I would understand if it's a blunder at a higher depth, and the computer says it's the best move at the default depth. But the computer knows it's a blunder at the default depth, even though it's supposed to be presenting the best line based on what it's calculated.