This may now be obvious from doc for precedence and doc for precedence classes.

So finally, the precedence relation is the one denoted by a precedence expression derived as follows:

The "top" LAT objects are those that no other LAT objects refer to with a *[object]*. Build a sequence, using (<precs> <precs>) as the sequencing operator, of all the top LAT objects, then repeatedly replace *[object]* instances by the precedence expressions in the objects they refer to.

A "class expression" is a precedence expression built using display addresses and the <df>=<df> operator. A class expression may be said to refer not only to those display addresses mentioned explicitly in it, but also to all those in the same precedence class. Replace the maximal class expressions in our precedence expression by class expressions that explicitly mention all and only those display forms in the precedence class referred to by the expression. (This is "normalizing" the form of class expressions.)

Finally, find normal form class expressions for all the classes not mentioned in our precedence expression so far, and add them to our expression by sequencing again with (<precs> <precs>).

The resulting precedence expression is of the simple kind described earlier when explaining which partial orders are possible precedence relations. IF YOU CAN SEE THIS go to /sfa/Nuprl/Shared/Xindentation_hack_doc.html