In a number of metrics, you will find the option to display the Rolling Average at the bottom of the metric settings. The Rolling Average creates a series of averages of different subsets of the full data set. It displays this average next to the actual data point to enable you to analyse trends in a more meaningful way (i.e. relative to recent trends and smoothening any peaks and troughs).
The Rolling Average in Plandek is dynamically based on the time range of the data you're querying as well as the granularity of the data displayed. In other words, the Rolling Average will vary depending on whether you're looking at 12 vs 4 weeks of data, and it will vary again depending on whether the data is plotted as weekly or daily values.
This is due to how we determine how "far back" to look in the data when calculating the rolling average window. A 12-week range of data that is displayed on a weekly basis will have 13 total data points (1 for each of the past 12 weeks plus 1 for the current week). The Rolling Average window is calculated by taking 20% of the total data points. Using the example above, that would be 20% of 13 data points, or 2.6 weeks. Given partial weeks aren't particularly useful for this calculation, we round the number up to 3 weeks. (Again, remember the number of data points will vary depending on your date range and display granularity)
The Rolling Average is calculated by averaging the values from the historical window AND the "current" week/day, meaning that if you are calculating the rolling average for a particular week over time, it will average the values from that particular week as well as each week within the historical window.