@ResearchVoyager: Thanks for the comment! One reason I chose to tackle gaming was that there's a ton of structured data that hasn't been leveraged well. A lot of the stats and probability models that power Gamerloop could be extended to other industries. There's a ton of stuff coming down the pipe, but the way I think about things is in terms of data derivatives (1st order = achievements, 2nd order = difficulty / completion rates, 3rd order = sequence/need inference, 4th order = matching based on need, 5th order = coming soon...).
@IRB: Easy/Normal/Hard does use relative completion rates as a signal; there was a lot of experimenting to figure out what the right criteria should be.
In a nutshell: Easy = casual gamers, Normal = mid-core gamers, Hard = generally only hard-core gamers. These values can change as Gamerloop gathers more info about a game or the algorithms get tweaked (the end goal is to provide useful recommendations to people).