I had been working from some 1-1-1 MSEM syntax that requested residuals for x,m,&y variables. It didn't make sense to request an x variable, because it has no residual variance. That must be why Mplus is changing them to y variables. It looks like I should be able to leave that part of the syntax out. It also appears that Mplus estimates residuals for the m & y variables by default anyway, so they should be able to be left out.