The page that scored zero, and why that is the right answer
Example.com returns 21 words and scores zero for Content and Entity Strength. It is also a perfectly working page doing exactly its job. Both of those are true and the scale has to hold them at the same time.
Updated
Example.com reads 36 degrees overall and scores a flat zero for Content and Entity Strength, the only zero in the published dataset. It is also a perfectly functioning page, served over HTTPS in 102 milliseconds, doing precisely the job it was created to do. Cold is not the same as broken, and a scale that cannot hold both of those at once is not much of a scale.
Why it is in the dataset
Every measuring instrument needs a known point. Example.com is the closest thing the web has to one: reserved by the IANA for documentation, unchanged for years, 21 words of visible text, no structured data, no author, no citations, no dates, nothing.
It anchors the bottom of the scale with something anyone can verify. If a future change to this engine makes example.com read 60, the change is wrong and the reference point is how I find out.
What it actually scores
Technical Accessibility: 85. It is HTTPS, it answers directly with no redirect, it responds in 102 milliseconds, it declares a viewport, a language and a character set. On the physical read it outperforms several institutions in this set.
Search Visibility: 42. It has a title. It has one H1. Nothing is blocking it.
AI Discoverability: 21. Twenty-one words is below every content threshold there is.
Content and Entity Strength: 0. No depth, no declared publisher, no author, no sameAs, no Open Graph, no outbound links, no dates. Nothing in that category is present because nothing in that category applies.
The useful part
A visitor whose site reads 36 degrees is going to feel something about that, and what they feel should depend on what their site is.
If it is a real business page, 36 is a genuine problem and the four sub-readings will point at which kind. If it is a landing page with a form on it, 36 might be entirely correct and require nothing.
This is why the scale page says the number describes the page and you supply the context, and why the bands are worded as descriptions rather than grades. "Readable, but nothing is helping it" is a statement about a page. A C minus is a statement about you.
A note on the zero
I thought about flooring the sub-readings at some small number. A zero looks like an error, and there is an argument that a page which is served correctly and answers instantly has earned more than nothing.
I left it. The reading is not a judgement of the page's worth, it is a count of observable entity signals, and the count is genuinely zero. Adding a floor to make the output feel less harsh would be the first small lie, and those do not stay small. The scale is easier to defend with a real zero in it than with a comfortable one.
What it demonstrates about the other sixteen
Every reading in the dataset is built the same way as this one. Same signals, same weights, same arithmetic, published with the date it was taken.
Example.com is the cheapest possible demonstration that the numbers are produced rather than assigned. Take the reading yourself. You will get 36, because the page has not changed since 1992 and neither has the method.
Filed under: calibration, example, scale