The numbers we refuse to publish
SlopCheckr computes figures it will not show you. Here is the minimum sample it publishes at, which statistics are withheld, and why a bigger corpus would not fix one of them.
Every statistic on this site carries the number of sites it was measured across. That convention exists because the alternative is easy and tempting: compute something from a handful of pages, round it, and print it in a large typeface where nobody can check the denominator.
This post is about the figures our own generator computes and then declines to show.
The threshold
SlopCheckr publishes a corpus figure only when at least 20 sites matched the condition it describes. The current corpus holds 93 scanned sites.
The threshold is arbitrary in the way all thresholds are arbitrary. It was chosen before the figures were computed, which is the only property that matters. A minimum sample decided after seeing which numbers cleared it would be a decoration rather than a rule.
The figures that clear it are on the methodology page, each with its own denominator attached.
What is withheld, and why
Median slop score by site builder
Only 3 of 93 scanned sites carry a detectable AI site-builder signature, below the 20-site minimum this project publishes at. The generic framework signatures we detect more often (Tailwind, Next.js) identify a toolchain, not a builder, so a median grouped by them would not answer the question.
Share of sites using the nav → hero → features → testimonials → pricing → FAQ → CTA → footer ordering
The section classifier labels the large majority of detected sections generically as "features", so it cannot currently distinguish a testimonials band from a feature band reliably enough to assert an ordering. Publishing this figure requires a better classifier, not a bigger corpus.
The two cases fail for different reasons, and the difference is the interesting part.
The first is a sample problem. Very few of the scanned sites carry a signature that identifies the tool that built them, and grouping by the framework signatures we can detect reliably would answer a different question — Tailwind and Next.js tell you a toolchain, not an authoring process. More scanning fixes this one, eventually, if site builders keep leaving detectable traces.
The second is a measurement problem, and scanning ten thousand more sites would not touch it. Our section classifier labels most detected sections generically. Until it can reliably tell a testimonials band from a feature band, any claim about the canonical ordering of sections would be a statement about the classifier's confusion rather than about the web. That figure needs better code, and no amount of data substitutes for it.
Why publish the absence
Three reasons, in increasing order of how much they matter.
Refusing a figure is cheap for us and useful to a reader, because the withheld list is a map of where the tool is currently weak. Anyone deciding whether to trust a score benefits more from knowing what it cannot see than from another confident number.
It also constrains us later. A published reason is a commitment: when the section classifier improves, the withheld entry has to change or disappear, and anyone can check whether it did.
Most of all, the product's entire claim is that generic, unsourced confidence is detectable and worth removing. A tool that made that argument while printing statistics nobody could check would be demonstrating the problem it sells the fix for. We ran the scanner at our own marketing site and found exactly that, which is written up in the self-audit.
The one number that survives everything
Scored 0 (no generic patterns detected) to 100 (every dimension fires).
93 of 93 sites scanned · measured 31 Aug 2026
Most of the web scores low, which is the result we would expect and the one that makes the scale useful. A detector tuned until everything looks generic has stopped measuring anything.
The scoring dimensions, the detectors behind each one, and the full published set of corpus figures are on the methodology page.
Sources
- Median slop score across the corpus — 15/100, 93 of 93 sites scanned, last updated 31 Aug 2026
Every figure above is computed from the 93 sites SlopCheckr has scanned, by scripts/compute-corpus-stats.mjs. Figures the corpus cannot yet support are listed, with the reason, on the methodology page.