Look into weighted replay stats by date

[dtracers]

For player stats we should weight them by how old they are. Player skills Change dramatically over time even within days if they make a big change. Newer replays should be weighted more heavily in addition to any cutoff.

[twobackfromtheend]

While good in theory, I highly doubt people play enough for us to have enough games for "new replays" to estimate stats well.

If we were to do this, I'd strongly suggest first looking at the stat distributions, seeing how many replays are needed for a good estimate of the means, and seeing if that threshold is met by whatever weighting we choose to use.