We did have some knowledge about one particular test. We knew that this test took the top 100 websites snapshotted in time and loaded each page one at a time, repeating 10 times. Talos loaded the page, waited for the mozAfterPaint event, (a standard event which is fired when Firefox has painted the canvas for the webpage) and then recorded the time from loading the page to receiving this event. Looking at the 1000 data points produced from a single test run, there was no obvious pattern. Imagine boiling those 10,000 points down to a single number and tracking that number over time. What if we made CSS parsing faster, but image loading slower? How would we detect that? Would it be possible to see page 17 slow down if all 99 other pages remained the same? To showcase how the values were calculated in the original version of Talos, consider the following numbers.