I've tried to work through recent discussion over the long weekend (when most of us in the US have been happily ignoring it during Thanksgiving) and though it was only a few of you, there was quite a bit of discussion.  I have tried to distill the discussion into a few manageable issues. I may not have covered it all but I am focusing on two issues:

1. Intervals and ranges
2. unknown/questionable/uncertain/approximate

 I have split these into two messages. This message covers the first, intervals and ranges. 

We've established (I hope) what we mean by an interval, that it has a start and and end, though there may be variations on that theme, for example the end may be "unknown" or "open", and the start and end might not be explicitly named - there may be a start and duration, or a duration and end, but in either of those two cases there is implicitly a start and end. 

And an "interval" is not the same as a "range".....

[1667, 1668, 1670-1672]   .....

I consider this a range. (The has been some discusison that a range has to have consecutive values but I don't think that's even meaningful nor that there is any value to such a restriction.)  Note the semantics of this example (as stated on the draft spec) "One of the years 1667, 1668, 1670, 1671, 1672". Or more generally for the function: "Uncertain, but known to be one of a set." I consider this a range, the same as the range of a mathematical function, a set from which a single value is chosen.

On the othe hand ....

{1667,1668, 1670-1672}

" All of the years 1667, 1668, 1670, 1671, 1672" or more generally "multiple dates".   ......

is not a range.   

The use case cited for this is "multiple dates, components of a single event" e.g. " 2002-10-{03,05,09-12}: the 3rd, 5th, and 9th-12th October 2002.

I believe it is worth discussion whether this is a worthwhile use case and whether we should be representing multiple dates at all. I can certainly see an argument against it, although the particular use case "multiple dates, components of a single event", was explicitly submitted (though I would have to dig into the archives to tell you who submitted it).  The case of multiple publication dates (as has been discussed), I do not think is a valid use case for representing multiple dates. Multiple publication dates can be supported by defining the publication date field as repeatable.