[soulgalore]

Comparing one run against another made changes hard to see: run-to-run noise is often bigger than the change you look for, and a single slow run can look like a regression. When both HAR files hold several runs of the same URL, the page now compares all of them by default. A change counts only when a Mann-Whitney U test says it is significant and it is at least 2%; with 10 or more runs per side the p-values are also adjusted for the number of metrics, so fewer changes are significant by chance. Single runs are still one click away, and outlier runs are marked.

The answer comes first: a verdict and key metrics with the spread of every run. The timeline follows the sitespeed.io rendering view: both videos play together, milestones light up as they pass, and the filmstrip, visual progress and waterfalls follow one cursor. Every metric from the old page is still there, but unchanged ones are folded so the table shows what moved, and a test fails if a metric goes missing.

Explaining a regression usually means pointing at a request, so you can now draw lines, circles and text on the waterfalls, move them into place, and save what you see as an image or in the saved comparison. The start page explains what the tool does, says that files stay in the browser, and tells you that sitespeed.io and Browsertime HAR files give you more to compare.

Assisted-by: Claude Opus 5.5 noreply@anthropic.com