I read an amazing post over on ars technica (link: http://arstechnica.com/science/news/2010/03/the-software-brains-behind-the-particle-colliders.ars?utm_source=rss&utm_medium=rss&utm_campaign=rss) abvout the LHC, or as they are also known, the "particle colliders". Beyond just the pure scientific geek awesomeness, these instruments have the potential to collect more data than you can (or possibly should) store.
Actually, this problem has a lot in common with a BI system. There's so much granular detail available in the source systems that a designer has to decide how, and how much, to roll up the data. Whenver you do that, you lose fidelity, but in many cases that's OK. Take, for example, your car's speedometer. You don't actually need to track each and every point of speed as it happens. You only need to know that you're hovering around the speed limit at a certain point in time.
Since this is the way that humans percieve data, is there some lesson we should take in the design of data "flows" - and what implications does this have for new technologies like StreamInsight?