Thanks for that note. These are deepish waters, and you might be interested in recent research on how to represent partial temporal knowledge, often motivated by practical data-mining problems. One place to start would be Ligozat's book Qualitative Spatial and Temporal Reasoning (2012), Wiley-ISTE, ISBN 978-1-84821-252-7. As an aside I should mention that I'm not a fan of interval arithmetic; if done carefully it typically results in intervals so large as to be meaningless, and if done sloppily what's the point? Admittedly I am biased here as Stott Parker and I wrote a paper <http://dx.doi.org/10.1002/spe.637> based on random rounding, a quite-different approach that doesn't require a new programming language. Of course we're way off the deep end here, in terms of any practical change I'd propose for the tz code or data. Still, it's often helpful to stay in shouting range of the researchers.