We had this discussion briefly, actually, and I voted in favor of Level.
There are tough cases on both sides (does 40 ± 1 really lose to 41 ± 10) but this solution errs in favor of those that play more games.
I think that using the 99.1% uncertainty is probably a little extreme, and that's the issue. If it were say 1 StD instead, you could still encode the uncertainty of player skill, but it would be a much smaller factor, like ± 3-4 for most people instead of ±10.
What's really annoying for me though, is my uncertainty is pretty high, so I'd have been double better off using flat raw skill. Uh, Theory, did you just change the sheet? I can't see players raw skill and uncertainty any more for some reason... wanted to check what my seed would be if it was using that.