Sourceforge is currently down, so I'm posting this here in lieu of a bug report. Will transfer it to a ticket assigned to me later. --------- The remarks associated with att.global.rendition say Where both @rendition and @rend are supplied, the latter is understood to override or complement the former. This seems correct to me. The prose of section 1.3.1.1.3 "Rendition Indicators" says In the TEI scheme, it is possible to supply information about the appearance of elements within a source document in the following distinct ways: 1. One or more properties may be specified as the default for all elements of a given type, using the @render attribute to point to rendition elements ; 2. One or more properties may be specified for individual element occurrences, using the @rend attribute with any convenient set of one or more sequence-indeterminate tokens; 3. One or more properties may be specified for individual element occurrences, using the @rendition attribute to point to rendition elements; 4. One or more properties may be supplied explicitly for individual element occurrences, using the @style attribute. If the same property is specified in more than one of the above ways, the one with the highest number in the list above is understood to be applicable. At the very least, 2 & 3 should be reversed, and use of the @selector of <rendition> needs to be added to the list. But the idea that @rend has to be made of tokens (rather than a string) seems incorrect to me; and the idea that they must be "sequence indeterminate" is at best confusing, at worst hogwash. And why does @style override @rend? I think I'd prefer the reverse (but I'm not sure). Typo: extra space before semicolon in <item> #1.