Five checks, each out of 100 from weighted parts. A part passes for full points, needs work for half, fails for none. The score is the mean of the checks measured: a one-time scan measures Find, Act and Safe (the static score); a watched site adds Finish and Stays correct.
Can agents find what you offer?
On every scan
| Part | Points | Passes when |
|---|---|---|
| AI crawlers allowed | 30 | robots.txt does not block GPTBot, ClaudeBot, PerplexityBot and the others (warn when some are blocked) |
| Sitemap | 20 | a Sitemap: line in robots.txt, or /sitemap.xml answers XML |
| llms.txt | 20 | /llms.txt answers text that starts with a # title |
| Structured data | 20 | at least one JSON-LD block that parses |
| Title and description | 10 | <title> and <meta name="description"> present and non-empty |
Can agents act on it?
On every scan
| Part | Points | Passes when |
|---|---|---|
| Named controls | 35 | every link, button and field has an accessible name (warn up to 2 without, fail above) |
| Headings | 15 | exactly one h1 and no skipped levels |
| Landmarks | 15 | a <main> landmark |
| Machine interface | 15 | a linked service description, /.well-known/mcp.json or an OpenAPI file (warn when only llms.txt exists) |
| Language | 10 | <html lang> is set |
| Image text | 10 | every <img> has alt |
Do agents finish real tasks?
On watched, verified sites
| Part | Points | Passes when |
|---|---|---|
| Three tasks | 100 | written from your pages; real agents attempt each one and every answer is graded against the page |
Do they do it safely?
On every scan
| Part | Points | Passes when |
|---|---|---|
| Open to agents | 40 | the page answers 200 to an agent’s request: no 403, 429 or bot challenge |
| HTTPS | 30 | the final URL is HTTPS and plain HTTP redirects to it |
| No CAPTCHA on read | 30 | no CAPTCHA widget in the markup of a page that only reads |
Does it stay correct after each deploy?
On watched sites, from the second scan
| Part | Points | Passes when |
|---|---|---|
| No regression | 100 | nothing that passed on the last scan fails now |
The home page you give, then robots.txt, sitemap.xml, llms.txt, three machine-interface paths and a plain-HTTP probe, all as ThemeliorBot/1.0, with an 8-second timeout and a 2 MB cap each. Redirects are followed; a bot challenge is recorded, never bypassed. Private and reserved addresses are refused before any request leaves.
Every part shows what was found, specific enough to check by hand, and the change that turns it into a pass, written for your page. The report orders fixes by the points each one adds to the total; “Check my fixes” scans again and shows before and after.
The first scan of themelior.com, on 17 October 2026, scored 67 with six findings. Each fix shipped in the week after; the scan of 24 October measures 100. The same six are the most common findings across the web.
| Fix | Check | Points in its check |
|---|---|---|
| Sitemap | Find | +20 |
| llms.txt | Find | +20 |
| Structured data | Find | +20 |
| Meta description | Find | +10 |
| Heading outline | Act | +15 |
| Machine interface | Act | +15 |
The method is a contract: read the example report to see every part with its evidence.