I created dpg whores because WN8 became a farce; tank choice was as much a factor as skill in determining your rating. Since each tank influenced WN8 differently, comparing WN8 was comparing apples and oranges, unless both players played the exact same tanks for the same number of battles.
My proposed solution is to keep things simple: go back to looking at hard stats, primarily dpg. I realize there are flaws with this approach, but there is one thing you cannot deny: the best pub players in the game tend to have the highest dpgs.
Before noobmeter stopped working properly, it used to have a tank-by-tank ranking that I really liked. There is no bs with this approach. You have a top 50 dpg in a particular tank or you don't; comparisons are really easy to make between players playing the same tank. It may be hard to compare players who play different tanks and that's ok. It's not any different from comparing a goalkeeper to a striker in soccer (please don't take this comparison too seriously); it only makes sense to compare players who play the same tank.
For players who aren't in the top 50, I simply display the percentile that the player is at based on both the dpg for that particular tank. If you're in the 90th percentile, then your dpg is higher than 90% of the population.
dpg whores is really simple. First, I gathered per tank data (dpg, frags, wr, etc.) for every player on the NA server with more than 100 battles. With this data, you automatically get the top players in every tank. For the top 50 dpg list, I only pick players who have 100 and 150 or more battles. For recent dpg, anyone with over 35 battles is included.
To create the percentile-based ratings, I created an empirical cdf based on dpg to estimate the distribution the data followed. I filtered the data by only using stats from players with more than 50 battles.
Most tanks follow a gamma distribution, the parameters for which you can estimate using MLE.
And that's all there is to it. For tanks with insufficient data due to few players, the percentile based rating isn't particularly reliable.
For recent stats, roughly 100 of the most recent battles are counted; if the oldest session can be dropped without the recent battle count dropping below 100, then it is. Any session more than 90 days old is also dropped. These numbers are still subject to change. If you have a well-reasoned opinion on this, let me know!
Some things I'm thinking of implementing are graphs of the dpg distribution for each tank. I think a tool for clan recruitment may also be useful, something like the average dpg/winrate percentile for all "important" tanks tier 8 and above, possibly grouped by tank class. If you think you have a good idea, let me know!