So say I play 50k hands and win at 4bb/100 with a sd of 85bb/100, between what 2 numbers can I say that my actual winrate lies with say 90% confidence, or 80% confidence?

Am I right in saying that this calculator...

http://www.castrovalva.com/~la/win.htm

will not be able to do this? Or can it?

For example when I type in 50k hands, 4bb/100, sd of 85bb/100 and 90% confidence in that calculator it comes up with -2.25bb/100 and + 10.25bb/100. Now, does this mean that IF my true winrate is 4bb/100 I can be 90% confident that my results after 50k hands will be between -2.25 to +10.25bb/100. Or does it mean that IF I play 50k hands and win at 4bb/100 there is a 90% chance that my true winrate lies between -2.25 and +10.25bb/100?

There is no way to know where your true win-rate lies using inferential statistics (like confidence intervals) alone. So yes, in your "for example" it means the first one; not the second.

Hmm well then how does this...

http://www.castrovalva.com/~la/winlose.htm

calculator work then? This gives you a confidence % that your winrate is above 0. If this can tell you what the chances are that your winrate is above 0 then why cant you use this to do what I asked in the op?

I'm not quite sure what that calculator is doing (even with the equations listed at the bottom). But I suspect it just makes the common misinterpretation of a confidence interval. A 95% CI (for example), does not mean you are 95% confident your true win-rate lies in the those bounds. It means that if your true win rate is equal to your current win rate, 95% of your results will fall within those bounds. These aren't quite the same thing. However, empirically they are often pretty close (e.g. I have seen some simulation studies that demonstrate that a 95% CI captures the true mean about 90% of the time).

You type in your current winrate, standard deviation and number of hands and it tells you what the chances are that you're a winning player.

Yes, I realize WHAT YOU DO. But I do not know for sure how it comes up with that number. The machinery is a bit hidden imo (although it does list some equations at the bottom). If those are the standard CI equations, they are in a form I am not used to seeing them in. So again, I'm not quite sure what the program is doing.

Ok, now I see what this program does. And it does as I thought.

An equivalent method is this:

t = M / (SD * sqrt(1/N))

Then, examine the probability of such a t value by using the normal distribution function (which is fine so long as N is > ~30). In excel this is =normsdist()

This returns the probability that of obtaining a positive win-rate if the true win-rate were actually zero.

I entered a win-rate of 10, an SD of 25, and 17 hands. The program returned a confidence of 95.05%.

Using the method above, I obtained a confidence of 95.0451%. Thus, the methods appear to be equivalent. Which means that this calculator is no different than any typical confidence interval calculator. It still relies on the null hypothesis that your true win-rate is zero, and returns 1 minus the probability of obtaining your results if that in fact your true win-rate was zero. This is NOT the same as the probability that you are a winning player.

This calculator provides the probability of obtaining such data given a M=0 hypothesis [ p(D|H) ]. To determine the probability that you are a winning player, you need an analysis that gives the probability of the hypothesis given the data [ p(H|D ]. Or in other words, a Bayesian analysis.