Charles JonesFor a person, a slow site is annoying. They wait, or they come back later.
An assistant does not wait and does not come back. It is working inside a time budget on behalf of someone who asked a question, and when a site exceeds that budget the request is abandoned. The business does not get a poor result. It gets no result, and the assistant answers from a site that responded.
We know this because we built one
Our fleet crawler fetches sites the way an assistant does, under a 15-second budget. Across 194 crawlable business sites:
| verdict | count | share |
|---|---|---|
| Readable without JavaScript | 131 | 67.5% |
| Timed out | 3 | 1.5% |
| Returned nothing usable | 4 | 2.1% |
| TLS broken | 3 | 1.5% |
Source — fleet crawl v0, 194 crawlable URLs, 2026-08-28.
Three sites are a small number. What matters is what happened to them: they were not scored badly. They were dropped, and nothing about the result distinguishes them from a business that does not exist.
Where the measurement stops and reasoning begins
Measured: a fetcher on a time budget abandons slow sites, and ours did.
Reasoned, not measured: that assistants in general apply similar budgets and behave the same way. We have not instrumented a commercial assistant to prove it. It follows from how they must work — nobody waits thirty seconds for a chatbot — but it is an argument, not a number, and this page will not dress it as one.
Quad 100, and why it is the floor rather than the trophy
Core Web Vitals score four categories out of 100: performance, accessibility, best practices, SEO. Scoring 100 on all four — quad 100 — is unremarkable on a single hand-tuned landing page.
The insurance platform holds 100/100/100/100 on mobile across page families at 388,329 pages.
Source — PageSpeed Insights, junelifeinsurance.com, mobile strategy, 2026-08-08.
The number is not the achievement. Holding it across a third of a million pages is, because every page is generated and any regression is a regression everywhere at once.
Why it is a correctness question, not a polish question
Performance work usually gets scheduled after correctness work, on the reasoning that a slow correct page beats a fast wrong one.
That ordering breaks when the reader is software on a timer. A page that does not arrive inside the budget is not slow — it is absent, and absent is a correctness failure. It belongs with the things that must be true before shipping, not with the things that would be nice to improve.