I hate these kind of hacks. In this case, a good solution (because of bandwidth and everything) would be: "give me all the updates since 1 hour _before_ what I think is the most up-to-date update)", and that would cover clocks going backwards up to 1 hour. And the cost would be in bandwidth, but not too much for this app.
Coding around it by explicitly checking for the clock movement is much harder than it sounds (or that it needs to be) - without installing pytz, you don't have access to this info. And even if you do have all the info - someone can change the timezone between two runs and mess things up. Really, if relying on timestamps, request.utcnow (when it makes an appearance in the source) is the way to go, rather than trying to figure out the quirks of request.now and local timezones.
