As BE refuses to show us their data what is and what is not balanced and how closely it is balanced, we could simply do that ourselves. Here's the how: We count how many events of a certain category A happen. This could be kills, deaths, matches won, matches lost... The problem is that we measure only a sample of a larger statistical ensemble. That means that our result is not the true result. The true result is somewhere near our measured result. How near? For that we would need to calculate confidence-intervals and significance-levels and stuff like that. We skip that. Let's simply say: If we have counted n events, our result has an absolute error-bar of sqrt(n). That means, our relative error is 1/sqrt(n). -> The more we measure, the more we count, the more accurate our result is, because the error-bars approach zero. (For a more detailed explanation, read up on Poisson-statistics.) number of events in a category is: n +/- sqrt(n) So, for example, if we want to know whether Eldar are OP, we could simply count how many matches Eldar have won. Let's say, Eldar win 4 out of X matches. They win a percentage of 4/X, but the error-bar is 2/X. Let's say, Eldar win 16 out of Y matches. They win a percentage of 16/Y, but the error-bar is 4/Y. Let's say, Eldar win 100 out of Z matches. They win a percentage of 100/Z, but the error-bar is 10/Z. (Now all you need is to play more than 100 matches that involve Eldar.) If we want, for example, to know whether Bolter are OP, we could simply look at the kill-death-ratio of characters who use Bolter and compare them to the kill-death-ratios of other characters. Let's say, we count k kills and d deaths. To calculate the error of our measured kill-death-ratio, we just need a little bit of gaussian error-propagation. kill-death-ratio and its error is: k/d +/- sqrt(k/d² + k²/d³) I proposed this two years ago. I wonder how toxic this thread will get this time.