Doug: Yes, that's exactly right. We can always produce a conservative estimate for when a game was cancelled by looking at the finished times of the games that weren't cancelled with higher IDs. These finished times can't be before the start time of the cancelled game.
There's another edge case, which is that some CDs have no associated game ID (such as CDs where the member returned to the position before anyone took it over). As far as I'm able to tell, figuring out when these were is impossible, and moving forward will require a change to the way CDs are recorded.
Neither of these issues are major, though, and as I said earlier I think it's worth designing any reliability rating overhaul without thinking about the implementation details. This way we'll get a system that is sensible, rather than one that is restricted by the current limitations of the system.
Leth: Whatever change we make to RR hasn't yet been decided, though I think it's clear that some kind of time cutoff or decay is necessary.
Kremmen: What are the other pressing issues that you see that we're not addressing? Fixing RR is important because games with CDs are a constant frustration for members, and clearly the current system isn't working very well - it incorrectly reports reliable players as poor reliability, and still lets some recently unreliable players through. We're working hard to keep the site improving, and if there are pain points that I don't know about - well, I want to know about them.