Claude+
ChatGPTClaude measured and built the fixes. ChatGPT audited the same measurements without seeing Claude's conclusions, then reviewed this page in three rounds. Its last round's changes were applied but not re-reviewed. What it raised.
We built two fixes on a test copy of the site, and measured them
Live today vs the template with the fix
Today's live theme has the video switched off. This morning's had it on. On the template it is on, and the page is still only 12.1 MB, because the video loads when you scroll to it. The Lighthouse score did not improve: 15 on the original template, 5 after the fix, 5 on live. The runs ranged from 1 to 24, so score is too noisy here. Weight and requests are the steady numbers. Phone processing work and page jumps are still poor. The template is a copy from before today's update, so small details differ from live, like the breadcrumb.
Live with Boost vs the template with our own filter
Numbers are medians across Best Selling, Coastal, The Statement Collection and Top 20 Overall (2 runs each). The screenshot is Best Selling. Best Selling alone: score 22 to 35, 7.1 to 5.7 MB. The template is an older copy with other differences, so this does not prove Boost alone caused the change. The switch would be made on the current live theme, with a side-by-side review first.
What each source downloads
Megabytes downloaded on the product page, not each one's speed cost.
Approve what you want live
What ChatGPT agreed with, challenged and requested
| Question | Claude | ChatGPT | Result |
|---|---|---|---|
| Biggest cause | Autoplay videos | Same: 24.4 MB average on 9 framed pages | Agree |
| Boost download size | First said about 340 KB | 530 KB to 1.2 MB, from 3 servers | ChatGPT right. Fixed. |
| "Blank for 10 to 12 s" | First stated as fact | It swings run to run | ChatGPT right. Removed. |
| Starting apps later | First said no visible change | Riskiest fix, test each app | ChatGPT right. Marked riskiest. |
| "Tested on a phone" | First wording | It is a phone-sized lab test | ChatGPT right. Reworded. |
| No same-template test | Compared live to staging | Needs before and after on one template | Done: 36.7 to 12.1 MB |
| Our filter faster? | Yes, recommend | Yes in these runs. Cause not isolated. | Agree. Side by side first. |
- Google Lighthouse: phone-sized screen, slow 4G, slow processor. A lab test, not real shoppers.
- Product page: 3 runs each, alternating original, fixed and live. Collections: 8 runs each (2 per collection). We show medians; with 8 runs that can be a half point.
- The template is a copy from before today's noon update. Live changed after.
- We do not headline first-paint time. It swings too much between runs.
- Video check: network log only. Vimeo requests at page load went from 5 to 0. We did not capture playback, so autoplay is not shown here.
All 35 pages tested this morning, before the theme update
Scores out of 100. Green 90+, amber 50 to 89, red under 50. Six pages (Homepage, Out Of Office, Out Of Office Unframed, Junk In The Trunk, Como From Above, Best Selling) come from an earlier run whose source files were not kept, so they are unverified.