I went back and checked — it’s interesting it named three assumptions. I don’t think it’s technically wrong and I’d give credit for it. Basically it sounds like what it is saying is the instrument must be independent of the confounders and calling that exogeneity, and then saying it can’t cause the outcome directly and saying that that is exclusion. That sounds a lot more like the LATE interpretation where exogeneity is independence. By referencing the error term of a regression, I agree the third is superfluous, because once it’s independent of the error, it’s independent of the confounders and can’t cause it directly. But it still amazes me it’s parsing things like this.