Requirements

• Tag System. You have to be able to identify and link different pieces of content together in unique and potentially dynamic ways across a number of dimensions. Tagging systems, particularly those that can access and create lists of "other posts with similar tags," are essential for providing some much needed organization to projects that are probably quite complex. Tagging systems should provide some way of supporting multiple tag namespaces. Also operations affecting tags need to be really efficient, from the users and software's perspective, or else it won't work at realistic scales.
• Static Generation. Content needs to be statically generated. There are so many good reasons to have static repositories. It allows you to plan releases (which is good if you need to coordinate documentation releases with software releases) most documentation changes infrequently. The truth is this feature alone isn't so important, but static generation makes the next several features possible or easier.
• Development Builds. As you work on documentation, it's important to be able to see what the entire resource will look like when published. This is a mass-preview mode, if you will. The issue here, is that unlike some kinds of web-based publications, documentation often needs to be updated in batches, and it's useful to be able to see those changes all at once because the can all interact in peculiar ways. These test builds need to be possible locally, so that work isn't dependent on a network connection, or shared infrastructure.
• Verification and Testing. While building "self-testing" documentation is really quite difficult (see also /technical-writing/dexy,) I think publication systems should be able to do "run tests" against documents and provide reports, even if the tests are just to make sure that new of software versions haven't been released, or that links still work. It's probably also a good idea to be able to verify that certain conventions are followed in terms of formatting: trailing white space, optional link formats, tagging conventions, required metadata, and so forth.
• Iteration Support. Documents need to be released and revised on various schedules as new versions and products are developed. Compounding this problem, old documentation (sometimes,) needs to hang around for backwards compatibility and legacy support. Document systems need to have flexible ways to tag documents as out of date, or establish pointers that say "the new version of this document is located here." It's possible to build this off of the tag system, but it's probably better for it to be a separate piece of data.
• Version Control. These systems are great for storing content, facilitating easy collaboration, and supporting parallel work. Diffs are a great way to provide feedback for writers, and having history is useful for being able to recreate and trace your past thinking when you have to revisit a document or decision weeks and months later.
• Lightweight Markup. It's dumb to make people write pure XML in pretty much every case. With rst, markdown, and pandoc the like there's no reason to write XML. Ever. of story End.
• Renaming and Reorganization. As document repositories grow and develop, it seems inevitable that the initial sketch of the organization for the body of work change. Documents will need to be moved, URLs will need to be redirected or rewritten, and links will need to be updated. The software needs to support this directly.
• Workflow Support. Documentation systems need to be able to facilitate editorial workflows and reviews. This should grow out of some combination of a private tag name space and a reporting feature for contributions, which can generate lists of pages to help groups distribute labor and effort.

This might just be a quirk of my approach, but I tend approach documentation, terms of process and tooling, as if it were programming and writing software. They aren't identical tasks, of course, but there are a lot of functional similarities And definitely enough to take advantage of the tooling and advances (i.e. make, git, etc.) that programmers have been able to build for themselves. Am I missing something or totally off base?