Rescue

What it proves — 6 claims

• The same shop, built twice, and the gap is a measured number rather than a claim. Its main content appeared after 8.4 seconds on the original and after 0.9 on the rebuild — nine times faster — timed on a mid-range phone profile over a throttled connection, three runs of each, with the raw reports kept.
• The two halves really are the same page. Both are built from one copy of the content, and a check refuses the build if their headings, image descriptions or structure ever drift apart. Without it, one side quietly changes and every figure becomes a comparison of two different pages.
• The slow half is slow in ordinary ways, not staged ones. Photographs at camera resolution, a carousel and its scripts loading before anything can appear, a web font arriving late, two utility libraries pulled in whole for one line each. There is no artificial delay anywhere in it — it is the shape most sites genuinely arrive in.
• The comparison happens in front of you. Both versions load on the one page and report their own timings as they go, so the figures are the ones your device just produced rather than screenshots of ours — and the page says plainly which measurements it cannot take that way.
• The heaviest thing on the original was a 3.5 MB video looping behind the header. The rebuild replaces it with something drawn on the fly that weighs a few kilobytes, and shows that measurement rather than asserting the video must have been worse.
• It is set up so the result cannot quietly slide back. The rebuilt page is re-measured against the same targets whenever it changes, which is the difference between a fast launch and a site that is still fast a year later.