Methodology · accuracy tests

We ran the scanner on the real web — and showed our work.

An honesty tool should hold itself to its own standard. So rather than assert an accuracy number, we ran reproducible tests across several real-world samples — the most-visited sites on the planet, and thousands of ordinary local businesses and online stores — each scanned with the free mechanical scan, the exact engine in the extension. Here's every method, funnel, and result, so you can check it or reproduce it yourself.

~71%

of the web's top-million elite scanned clean

~29%

surfaced at least one flag worth a look

~10%

carried at least one HIGH-severity flag

Top-million sample: 49,997 sites drawn at random across the top 1M (~50k). Ruleset v2026.07.08 · scanned 2026-07-09.

Even this is the elite. A random draw from the top million is still the top ~0.1–0.5% of the web by traffic — the most-visited, best-resourced sites, not "the average website." The hundreds of millions of smaller sites below the top million weren't measured here; if anything, the trade data further down suggests they'd fare worse, not better.

The one finding that matters: these flags are everywhere. Whether we looked at the world's most-visited sites or ordinary Main-Street businesses, roughly 3 in 10 tripped at least one pattern regulators are on record scrutinizing. It isn't a fringe problem or a "sleazy niche" problem — it's a default-state problem. And in some trades it's far worse (see below).

Four samples, one engine

Each row is an independent, separately-labeled test. Same scanner (collectSnapshot + ruleset.evaluate), same honest funnel (a site counts only if it's reachable, has real English body copy, and could actually be evaluated).

Sample	Sourced from	Scanned	Clean	≥1 flag	≥1 HIGH
Top-tier sites (the elite)	Tranco top-600	~600	~80%	~20%	~4%
Top-million (random)	Tranco top-1M, random draw	49,997	70.7%	29.3%	10.5%
Local businesses	OpenStreetMap, 5 US metros	8,311	69.6%	30.4%	7.8%
Online stores	Common Crawl (Shopify)	3,135	80.2%	19.8%	6.2%

Read the "online stores" row carefully. It's sourced from *.myshopify.com subdomain stores, which skew toward newer / under-built shops (established brands sit on custom domains we can't enumerate here). Under-built shops have little copy to flag, so this number reads artificially clean — treat it as a floor for one narrow slice, not as "ecommerce in general."

The part that should get your attention: it varies wildly by trade

The 8,311 local businesses weren't uniform. Broken out by trade, some verticals are flag magnets — and these are ordinary, well-meaning small businesses, not bad actors:

Trade	≥1 flag	≥1 HIGH	n
Home & trades (plumbers, roofers, HVAC, cleaning)	56.0%	22.0%	455
Auto (dealers, repair)	47.4%	20.3%	443
Retail shops	35.4%	6.8%	1,821
Beauty & personal care	31.7%	11.4%	518
Professional services	31.2%	5.9%	1,348
Fitness & recreation	30.4%	10.8%	526
Food & drink	29.7%	1.9%	1,785
Health & medical	21.2%	8.5%	986
Education	13.0%	5.6%	391

Individual trades sharpen it further (sites with enough sample to be meaningful): car dealerships 58% flagged / 31% HIGH, self-storage 72% / 24%, gyms 48% / 16%, plumbers 49% / 16%, furniture stores 43% / 21%, auto repair 42% / 14%. The pattern: trades that lean on urgency, guarantees, and "#1 / best in town" language light up — exactly the phrases with a real enforcement history.

How each sample was drawn

Top-million — random draw from a public ranking

Source: Tranco top-1,000,000 (research-standard, manipulation-resistant), list ID 46XZX (2026-07-08), archived SHA-256 1bdc9757…54158c. List ↗
Draw: mulberry32 PRNG, seed 20260709, uniform over the full million. Sample-to-target with replacement on evaluability only, never on findings. No site chosen by us.

Local businesses — real businesses from OpenStreetMap

Source: OpenStreetMap (Overpass), businesses carrying a website tag, by OSM category (e.g. craft=plumber, shop=car, amenity=restaurant), across 5 US metros (NYC, LA, Chicago, Houston, Atlanta). 11,375 real sites collected across 123 niche categories.
Note: Metro-bounded, so US-centric; a niche's number reflects those metros. Categories map from the 37 rule categories.

Online stores — Shopify via Common Crawl

Source: Common Crawl index CC-MAIN-2026-25, all *.myshopify.com hosts (12,730 unique), seeded random sample. Reproducible from the crawl ID + seed. (See the skew caveat above.)

What these numbers are — and aren't

Mechanical scan, pre-AI. Pattern matches, tuned for precision but willing to raise a hand — so a flagged rate is an upper bound on real issues, and the optional AI review drops matches that don't hold up in context. "Flagged" = worth a second look, never "breaking the law."
Most flags are low-severity. The common ones are soft "worth a look" patterns (a "#1" claim, a countdown timer); the genuinely serious HIGH-severity flags are rarer. A flag is a prompt to check, not a verdict.
English-language sites only. The ruleset is English; non-English pages are excluded from the denominator, not scored.
Small-sample trades wobble. Trade breakouts with small n (roofer n=23, tanning n=23) have wide error bars — trust the group totals and the higher-n trades.
A snapshot in time. Sites change, enforcement evolves, the ruleset is versioned (v2026.07.08). Re-running later will move the numbers.

Reproduce it

The sources are public and the draws are seeded. Full harnesses, sourced lists, and per-site results live in the project's research/ folder; the top-million list is Tranco 46XZX (seed 20260709), the Shopify set is Common Crawl CC-MAIN-2026-25 (*.myshopify.com), and the local set is OpenStreetMap by category. Same inputs → same numbers.