About this studio
Pikkola, tool No. 20
One request from the edge for what the visitor waits for, one read of the head for what the crawler sees, and three files written for the parts that are missing.
Why not a browser
A browser tells you how a page felt on your laptop, on your connection, with your extensions. That number is useful for you and misleading for everyone else. This studio takes the other half of the picture: what the server sends, how long it takes to start sending it, and what a crawler would find in the response. It answers in about a second, from any device, without installing anything.
The trade is real and worth saying out loud. There is no Largest Contentful Paint here, no layout shift, no interaction delay, and no field data from real visitors. Those need a browser and real traffic. What is measured here is the floor: the part of the experience that the response itself decides.
How the two scores are weighed
Search readiness carries 100 points across thirteen checks: title 12, description 12, canonical 8, one H1 8, heading outline 6, image alt 10, indexable 8, language 4, viewport 5, social tags 7, structured data 5, robots.txt 8 and sitemap.xml 7. A check earns full marks, half marks or nothing; the score is the share earned.
Load cost carries 100 points across eight checks: time to first byte 30, compression 15, HTML weight 15, blocking scripts 15, blocking stylesheets 10, resource hints 5, inline payload 5 and third party scripts 5. Time to first byte is scored at 500 ms or less for full marks, 1000 ms for half, above that for none.
The overall score is the two halves equally, and the letter is a band of that overall: A+ from 95, A from 85, B from 75, C from 65, D from 50. The letter is a summary, not a verdict. A page can be fast and invisible, or slow and perfectly indexed, and the two numbers say which of those you have.
Reading the crawl files
robots.txt is fetched once. The parser looks for a wildcard group that disallows everything, a crawl delay and any Sitemap lines, and counts the disallow rules. It is not a full RFC 9309 implementation: it reads the directives this studio acts on, and treats anything else as text.
sitemap.xml is looked for at the conventional path first, then wherever robots.txt points, then at /sitemap_index.xml, then in a link element on the page. The first one that answers is the one reported. URL counts come from <loc> elements, and a <sitemapindex> root is reported as an index rather than a list of pages.
When one of them is missing, the generated file is a starting point, not a policy. The robots file removes admin and api paths because that is a common shape, and the sitemap lists the page plus the internal links found in the HTML, up to twenty. Both need your judgement before they are published.
The private address gate
A server that fetches any address it is handed can be pointed at a router page, an internal console or a cloud metadata endpoint. Every host is resolved through DNS-over-HTTPS before the request and refused if the answer lands in loopback, link local, a private range, carrier grade NAT, multicast or a metadata address. Names that only exist inside a network are refused before resolution, and each redirect target and both crawl files go through the same gate.
This raises the cost of abuse and does not remove it: a DNS answer can change between the check and the request. That is why the worker only opens ordinary web ports, reads only the HTML and the two crawl files, and never returns a body from any other content type.
What it does not do
It does not run JavaScript, so anything rendered on the client is invisible to every check here. It does not test forms, logins, TLS versions or certificate chains. It does not crawl beyond the page you give it, and it does not keep a history: every audit starts from nothing, which is also why the same address can score differently on two days without anything having changed.
It also refuses to audit itself. A worker cannot fetch its own route: the subrequest would go straight to the Pages origin, so the answer would describe a page no visitor receives. Saying so is more useful than printing a grade for the wrong layer.
Who runs it
Pikkola Lab builds small, free, single purpose tools and a few paid ones. This studio is free, has no account, stores nothing, and is paid for by the subscription next to it. Questions go to hello@pikkola.com.