web10WebPagetest parameter.) One result of this is that the number of requests and total size of pages increased because more network requests were captured.
Article published on CNET News citing various statistics from the HTTP Archive.
The first three runs in the HTTP Archive (Oct 5, Oct 22, and Nov 6) only had ~1000 URLs. Comparing those small runs to the later larger runs (that have ~17,000 URLs) can be misleading. (See this blog post.) This change removes those first three runs from the UI to avoid these confusing comparisons. The data is still downloadable.
We used to run har_to_pagespeed to generate Page Speed scores. With this change the Page Speed scores are generated from WebPagetest. This means that more rules are evaluated and the score is more informative. When using har_to_pagespeed the input is a HAR file so any rules requiring access to the page's DOM can not be scored. WebPagetest, however, generates the Page Speed score while the page's DOM is available, thus resulting in a more accurate evaluate of the page's performance.
One side effect of this change is that comparisons of Page Speed scores to runs performed before March 25 are not apples-to-apples comparisons because this change in which rules are evaluated.