WebPagetest/AFT FAQ

Life's too short for slow pages

Why doesn't AFT work for 100% of the urls? It is not always possible to distinguish animation from frame flow using only screenshot analysis. For example, think about the boxes that flash on the screen when you load bing.com. Think also about the map widgets, which sometimes flash on the screen as part of the page loading process. We know from our experience using these sites that in one case, the flashing icon is an animation meant to draw our attention to something of interest, but in the other is the result of frame flow. Restricted to only a series of screenshots showing each page loading, it is not possible to distinguish whether the flashing is intentional.

What does the 'high confidence' status message indicate? AFT doesn't work 100% of the time, but when animation can't be distinguished from frame flow, the status message will indicate that this is the case. The false positive rate is very low. I have never seen a result marked as 'high confidence' for which the AFT result was not what would be expected based on manual inspection of the screenshots. The 'high confidence' results are extremely reliable.

What percentage of AFT results are 'high confidence'? This depends on the set of sites you are using, but on a random sampling of Alexa top sites, over 80% returned 'high confidence' AFT results.

What should I do to measure AFT latency for results that do not return 'high confidence' results? There are several options. Urls in this category either have a one time animation that is long, or an unusually large amount of frame flow. The parameters can be adjusted to either take the point at which the long animation/frame flow stops, or to pick a point at which a significant portion of the page is loaded, such as latest first render.

What does AFT measure? The time from the start of the page load until when content that stops changing and is above-the-fold reaches its final state.