The same 100 businesses, measured every few days.

They are kept in the record exactly as measured. They were taken on an earlier list, by an older version of the checking tool, or with our reference pages counted in.

Every standing run through October 5, 2026 counted our two reference pages inside its figures. The record keeps only totals, so those figures cannot be fully separated after the fact.

Before this standing list existed, a one-time study tried 100 sites and graded 71 to set the starting point: middle score 82. It used a different list of sites, so it is shown next to the table, never inside it.

Every run also reads two pages that are not census businesses and are not in the list. One is a large company's carefully built home page. The other is a nearly empty test page. We read them so that a shift in our own checking tool shows up. If their scores move, we suspect the tool before any website.

Since October 12, 2026, their scores are kept on a line of their own and never counted in the figures on this page. On the latest run the carefully built page scored 92 and the nearly empty page 68.

We do not name either page; this page names no site it measures.

Each had been graded on every visit we made, on at least three different days, and read within the 14 days before the draw. We grouped them by kind of business and by Florida or elsewhere, in the census's own shares. Inside each group a fixed shuffle chose them, so no person picked a site.

13 of the 100 sites did not answer our checking tool with a real page: a screen asking us to prove we are a person, no answer in time, or a refusal. Grading that screen would be grading the lock on the door, and grading a silence would be grading the weather, so the tool refuses to score what it cannot honestly see.

That honesty gate applies to every site, ours included.

When the middle score changes between two runs, it may reflect sites that improved or declined. It may also reflect which sites answered: a site that times out one run and responds the next changes the middle score even if nothing on the site changed. Refusals are excluded from the graded count shown in the table, so a run that reaches more sites than the last one is comparing a different set. The Graded column shows how many sites each number describes.

The record behind this page stores middle scores, how widely scores vary, grade counts and how often each check failed. It stores no site names next to scores and no per-site rows at all, so nobody can use it to shame a specific business.

Check names are public, and since 2026-09-15 so is what each check is worth: the full table is on the proof page. For most checks, we do not publish the exact cutoff between pass and fail.

This page reports one standing list of 100 sites, re-measured on a schedule so its numbers can be compared over time. Each version of the checking tool gets its own line. Separately we have read a far larger collection of business websites, kept so a site can be placed among its own trade rather than against a general standard. The two are different instruments and their numbers are never added together or averaged into one line, which is why they get separate pages. The Three Rings is the record of that wider collection.

The same checking tool behind this benchmark runs the free inspection. Same checks, same honesty gate, same refusal to guess.