I don't think it was ever precisely accurate in absolute terms, and it surely isn't more accurate now, but it appears to be accurate in relative terms -- i.e., as percentage changes in activity over time. A semi-random sampling of subreddits corroborates the conclusions of the data (that there are far fewer user contributions now).

makes sense. if it was consistently inaccurate, the drop in traffic relative to the moving average would be captured.