Section #SACR discusses encoding and processing of canonical references. However, there are some suspicious statements:
1) "When an application encounters a canonical reference as the value of cRef attribute, it follows a sequence of specific steps to transform it into a URI reference." -- We seem to be describing how we think hypothetical software operates. I suggest changing "it follows" to "it might follow" or something less cumbersome like "these Guidelines recommend that it follow".
2) "It is an error to reference more matched substrings than are produced by the regular expression." -- Unless one or more of the various schema langauges generated by Roma/Byzantium can detect this, it seems we mean to say that it is not TEI-conformant to encode a document in this way. I suggest changing to say "It is not conformant to these Guidelines to reference [. . .]". Also adjust the wording in the following sentence ("would produce an error").