Thanks for the comments guys! I will revise as follows:
1. I will allow omission of time
2. I will change the format to require dashes to separate fields in the date and colons in the time
3. I will permit either dash or slash to separate documentname and datetimestamp
4. I will use a T to separate date from time. QUESTION to commenters - should "T" be separated from informational fields by dashes on both sides?
5. I will support coarser time
6. I will NOT support ordinal date; the rationale here is that if the same document is annotated two ways (once in one archive and once in another) it will not be easy to tell that the documents are the same, as their datetimestamps will be lexically different.
7. The specification already says that datetimestamps are UTC; I will revise to require the field to end with Z to indicate this.

In my background, the hard part is not creating urls for documents that already have citations formats (though that is valuable), the hard part is creating citation formats for content whose citations include page numbers.

In general, citations (more broadly speaking than URIs) should have the following properties:
* vendor (ie publisher)-neutral
* medium neutral (I shouldn't have to have a specific publisher's book to find the content, and I should also be able to find the content in an old-fashioned library if I am offline)
* public domain (nobody should own the citations - this has been a huge issue, believe or not in legal publishing)

A URI (URL)-scheme seems like it could address many of these - but in print, how would this scheme be used? Would things like bluebook citations be replaced by URLs? Would I then have URLs sprinkled all over govt documents (in print)? (That is not out of the question, btw -- its just that its a culture shock).

I guess my point is that while we are happy to see stuff moving online with law.gov, lets not forget that there is a huuuuuge world of dead-tree users out there and there are valid reasons to support their use cases as well.

I think if something was first distributed as a book - it may be easier to create a URL w somethign similar to a book annotation and hide the internal references. But UGH what a difficult site that would be to read! I think most of these are posted as PDF's and maybe that part should go be included on the PDF page?

But for a website, I see no reason to make it more complex than pagename datetimestamp and paragraph

Tantek: I believe I've now addressed your comments.

Gabe: Page numbers are formatting artifacts rather than structural elements of a document. My objective here has been to create a format which will allow a document's structural sections to be referenced directly, without requiring recourse to metadata such as page numbers (which vary according to page size etc...) This format is intended to be medium-neutral in the sense that the display medium doesn't modify the document's underlying structure & so references of the type we define should be preserved even when the display medium changes. Documents which are natively Citable in compliance with this specification don't require index metadata, and so don't create a requirement or opportunity for proprietary third-party metadata indexes like those currently in use in legal publishing. Citations in print should be to chapter and verse (in other words, "Section 1, Subsection a", and the chapter and verse headings should be printed as part of the document itself. There could be an automatic translation from this format to a canonical Citable Document Subdivision Locator format for particular reference structures, but it's not clear that what's suitable for dead-tree reference translates automatically to electronic reference or vice-versa.