CrawlCheck

Findings · 2026-08-20 · By

Why caching only successes kept our scans slow

Scans of small business sites ran slow and varied from run to run, because we kept asking a question whose answer was always no. Our first fix made one case worse.

Scans of small business sites were slower than they needed to be, and the same site could take nearly three times longer on one run than another. The cause: a lookup for public real-world performance data, which small sites almost never have, remembered its answer only when data came back. A "no data" answer was asked again on every scan, costing up to 8.2 seconds. Remembering the "no" cut it to 3 ms. The rule: cache facts about what you measured, including "nothing there", never your own failed attempts.

The repeated “no data” lookup, before and after
Beforeup to 8.2 s on every scan
Once the “no” was remembered3 ms

Three scans of the same site, with no change in between, took 12.9 seconds, 20.6 seconds, and nearly three times the fastest. An average would have sent us hunting for something twenty seconds long. There was nothing twenty seconds long: there was a fast path and a slow path, and averaging them hid both.

What was wrong? #

Part of a scan asks whether public real-world performance data exists for the site. For most local businesses it does not, and it will not: the Chrome UX Report only publishes a site once it has enough Chrome visits, and a regional business rarely gets them. We remembered the answer only when data came back. A "no" was never remembered, so every scan of a small site paid again for the same empty answer.

MeasureValue
Fastest of three runs, same site12.9 s
Middle run20.6 s
Slowest runnearly three times the fastest
Repeated "no data" lookup, beforeup to 8.2 s on every scan
Same lookup, once the "no" was remembered3 ms
Side effect of our first fixan intermittent failure frozen for six hours

How did our first fix make it worse? #

The first fix remembered every outcome, including a temporary "still waiting" state. That is not an answer about the site at all; it is our own request not having finished. Remembering it stopped the retry from running, and an intermittent failure became a permanent one for six hours until we corrected it. The scan results themselves never changed: the measurement was the same, it just stopped repeating an empty lookup.

What carries over to your own tools? #

  1. Time each step, not the total. A total cannot tell one slow step from one occasionally slow step, and they need opposite fixes. The same trap applies to single Lighthouse runs.
  2. Remember "the answer is no". It is a fact about the thing you measured and stays true as long as a yes would.
  3. Never remember "we could not ask". A rejected key or a request still pending is a fact about you. Retry it.

If your own site is one of the many with no Chrome UX Report data, here is how to measure Core Web Vitals without it.

CrawlCheck’s dataset when this page loaded: 582,334 sites measured.

Share this finding:LinkedInXEmailFollow Emmanuel on LinkedIn

Questions this post answers

What is a negative cache?

A stored record that a lookup found nothing, with its own expiry, so the same empty request is not repeated on every run. A cache that only stores successes repeats the expensive miss forever.

Why do small business sites have no Chrome UX Report data?

The Chrome UX Report only publishes a site once it has enough real Chrome visits, and a regional business rarely reaches that level.

Did this change any scan result?

No. The measurement was the same before and after. Scans of small sites simply stopped repeating a lookup whose answer was already known.

What should never be cached?

Anything that describes your own attempt rather than the thing you measured, such as a rejected credential or a request that has not finished. Those must be retried.

Related findings

The main product

Found this on your own site? We fix it for $100.

Scan free to see where you stand. The fix is one site, every finding implemented and re-measured, with a sealed before and after.

Free browser extension

Check this on any site you open, without leaving the page.

CrawlCheck for Chrome grades every site you visit A–F and shows, from your own browser, what search engines and AI assistants are actually served. Free, no account.

Write for the library

Measured something yourself? Submit an article.

Send your finding, the data and the method. A person reads every submission and replies by email; accepted articles are published under your name with a link to your study. Free to submit. Articles are accepted on merit; no one can pay to have one published.

Submit an article

How it works · a reply within about a week

Spotted a mistake or a bug on this page? Report an error. Confirmed errors are fixed and logged on the corrections page.

Comments

Comments are read before they appear. Nothing is published automatically, and no account is needed. An AI agent with a published signing key can post without the queue: it signs the request and the signature is checked against its own domain (how). Those appear at once, marked verified.

Writing about this? Facts, live figures and marks — every number on that page is dated and traceable to a scan.

All findings · The dataset · How the dataset works