A precondition is to have a fairly complete and explicit set of rules about which markup is acceptable, necessary, and sufficient (which may if necessary differ by project, but differences should be avoided.) The accomplishment of the application of the necessary and sufficient markup should produce a single version which would comprise a canonical source for all final published versions by the application of stylesheets, translators, XSLT transformations, unix-style filters, manual adjustments and enhancements, or other means. (Additional resources may require inclusion, such as images for illustrations, etc.)
I don't see where this is any different than the "rules" which are in place for HTML right now. PG says what it wants, DP submits something else, and PG accepts what DP offers because PG wants what DP offers. Then Marcello tries to fix up that which was offered so it can run on something smaller than a 20" computer monitor. If you can get DP to agree to submit a flavor of "HTML" or "XML" or the like which is closer to their internal informal formatting markup language, and less page-image-formatting-oriented than their finally submitted HTML, then you would be ahead in the game. In that case, you are basically co-opting that critical mass of people already using a somewhat-reasonable markup language internal to DP to try to generate agreement about what should be in or out of that markup language, and what the syntax of that markup language should be. And DP is in a much better situation to enforce those rules than the people active on this forum. http://www.pgdp.net/c/faq/document.php Unfortunately, what DP is doing right now in their internal-format formatting files is a combination of troff-style formatting and XML markup. It would be much easier to create tools if it was all just 100% XML-style markup. People thinking about this issue would do well to look at their set of agreed-upon formatting markup rules, which literally contains about 100 markup rules. Agreeing on a formatting standard is NOT a simple task.