Srry for bumping this but it's a good post so w/e.
In fact, masterLJ, variance is even sicker than you describe.
Let's say you have a 10ptbb winrate over 100k hands. You might think the chances of your true winrate being 5ptbb or lower are equal to your true winrate being 15ptbb or higher, but that's not true.
The population of winrates isn't normally distributed, there are much more true winrates of 7ptbb then there are of 13ptbb. So, when you have a 10ptbb winrate over 100k hands, your true winrate is indeed between [10-x,10+x] with a certain confidence (obviously the confidence you want, together with your SD, determines x), but it's not symmetrical*! You have higher chance of your true winrate being [-infinity, 10] than of your true winrate being [10,infinity].
Unfortunately, we don't know how true winrates are distributed. We know it's 'weighted downwards': 7pbb winrates are more likely than 13ptbb winrates. But we don't know how much this effect plays (the higher your observed winrate it the more it plays though, for obvious reasons). In addition to that, you will get better. I guess the point is when you have an observed winrate that you are crushing the game, not only do you have the standard confidence interval putting your two feets in the ground, there is also another effect in play, making it even more likely you are 'just' a good player running hot instead of the next durrr (becausere there are more good players than durrrrrs)
English isn't my first language so it's kinda hard to put my thoughts into words, I'll just refer to mathematics of poker chapter 3 (page 40-43 in particular)
*the chance of the results of the next 100 hand sample being either [-inf,9] or [11,inf] is the same/symmetrical though