[age_bronze]

I don't think they correctly realized the meaning of these blunders, most games are skilled-vs-skilled or amateur-vs-amateur, and for every blunder a skilled person makes there was a trap another skilled player placed. The way I see it, skilled players are more experienced at leading their opponents into exploitable complex conditions, whereas amateurs tend towards simpler board positions that are easier to reason about, and it is not uncommon for blunders to be missed by both sides.

Anyway I'm not sure they could control for the skill of the players in both sets, because their dataset is probably missing amateur-vs-skilled data, which probably renders most of their skill-based correlations incomplete, as we don't know if they correlate to the skill of the player making the mistake or their opponent.