The system was never very unstable. The error rates are all low. The point is that it's now much more stable.

The reason I didn't sort error rates by month was to preserve statistical significance. If we try to only look at error rates from specific periods of time, we have to decide how short those periods of time are. Too long and we don't have data from some of the browsers; too short and we don't have enough data to form a significant opinion. It may have been possible to find a middle ground, but why bother when we can still get unbiased results for 80% of the browsers?