It took me longer than I want to admit here to mentally parse the example at the end of section 8.3.1 on \mathstyle, \mathchoice, and grouping. In particular it took a while until I noticed that the b case has no \bgroup ... \egroup. This may very well be because of temporary stupidity on my side. But nonetheless I recommend changing the example either so that the mathchoices between \bgroup ... \egroup receive one more level of indentation, or, even better, to change the example into something like: \def\a#1{ \mathchoice {\bf \scriptstyle (#1:d :\mathstyle)} {\bf \scriptscriptstyle (#1:t :\mathstyle)} {\bf \scriptscriptstyle (#1:s :\mathstyle)} {\bf \scriptscriptstyle (#1:ss:\mathstyle)}} [a:\mathstyle]\quad \bgroup\a x\egroup \quad [b:\mathstyle]\quad \a y \quad [c:\mathstyle]\quad \bgroup\a z\egroup \quad [d:\mathstyle] Apart from being shorter, that totally minimizes the effort required by the reader to mentally parse and compare what is equal and what is different in the example.