# What a median score by sector tells you, and what a global ranking does not

The median local business scores 59. The median SaaS company scores 74. A single leaderboard across both mostly measures which kind of business a site is.

Published: 2026-09-09
Category: Method
Figures as of 9 September 2026, from 350 scanned sites.

In a sweep of 44 sites, the median score for a north-Seattle local business was 59.5. For SaaS companies it was 74. For a mixed sample of everything, 54.

That spread is wide enough that a single global ranking is largely a sector classifier with extra steps. Telling a bakery it ranks below Zapier is arithmetic, not information.

## Why the spread exists

Not because software companies try harder. Because the checks that separate a 74 from a 59 are things a software company gets for free: a framework that server-renders, developer documentation that exists anyway, structured data a marketing site template shipped with, an engineer who noticed the cache headers.

A dentist has none of that and does not need most of it. Their site can be completely legible to a reading agent and still score in the fifties against a catalog weighted for a category they are not in.

## What we compare against instead

The cohort is the scoring profile — the same thing that decided which checks the site was measured on. Two sites in the same profile have the same denominator, so their totals are on the same scale and a median across them means something. A median across mixed profiles averages numbers built from different check sets, which reads like a comparison and is not one.

> A profile is only ever selected from what a site declares about itself in JSON-LD, using schema.org's vocabulary. Never inferred from the absence of the thing being exempted, and never from `Product`, `Offer` or `WebPage` — a SaaS company declares all three, and matching on them filed every site with a price on it as a shop.

## When we say nothing at all

A median is only as good as what it was taken over, so the comparison refuses to appear in three cases:

- Fewer than twelve scored sites in the cohort. A median of three is an anecdote with a decimal point, and it is exactly the kind of number somebody would act on.
- A cohort of one, which is the site looking at itself. It would print "exactly the median", which reads as a compliment and is arithmetic over a single row.
- A scoring version with nothing scored under it yet — the normal state for a day or two after a version ships, because a median has to be computed within a version or it compares totals built from different catalogs.

And when it does appear, the count is in the sentence rather than in a footnote: "13 points above the median for a local business, taken from 44 scored sites." You are entitled to weigh it without hunting for the denominator.

## The general lesson

Any single number describing a website is an average over a choice of what to measure, and the choice does more work than the arithmetic. That is true of ours. It is why every weight we use is published, why every score records the version that produced it, and why the current medians and counts are live on [what we have measured](/stats) rather than quoted from the sweep that found them.

## Related

- [Nine in ten sites have no agent manifest, and most of them should not](/blog/nine-in-ten-sites-have-no-agent-manifest): A check that 90% of sites fail is not a finding, it is a wall. Here is why we rebuilt a whole scoring category around that, and what we changed it to.
- [Six pages, one second apart, and the things we refuse to do](/blog/six-pages-one-second-apart): What our crawler requests, what it never does, and why a scanner that could get past a block would produce a worse number than one that cannot.