About
A website is now judged by search engines, AI assistants, accessibility law, security expectations and performance budgets — five authorities, five rulebooks, and no agreement between them. Almost every audit tool was built to answer one. So you run four, export four reports, reconcile them by hand, and hope nothing fell through the gaps between them.
WebsiteValidator exists to be one assessment across all five, scored deterministically against published standards rather than opinion.
And an audit you can argue with. Most auditors hand you a number and ask you to trust it. This one hands you the rule that produced it, the evidence it matched, the specification it came from, and the date that rule was last checked against its source. If you think a finding is wrong, you have everything you need to prove it.
Every status and every point of impact comes from a deterministic check with a declared confidence tier and a citable reference. A language model can rephrase an explanation into friendlier prose; it is structurally incapable of changing a status or a score, because it is never given one to change.
A rule backed by the HTML specification and a rule backed by a blog post should not carry the same weight, and inventing a number like 0.73 to express that difference is false precision. Rules declare SPECIFICATION, RESEARCH, EMERGING or EXPERIMENTAL, and two rules with the same class of evidence always carry the same weight.
Asking whether a page declares Recipe markup is useful in an audit and meaningless as a penalty — a law firm is not defective for lacking a recipe. Roughly two thirds of the checks here are informational: they appear in the report and are excluded from the score entirely, so breadth never dilutes what the number means.
Rendering is a real browser. Core Web Vitals are measured on the live page. Accessibility runs industry-standard WCAG rule engines against the rendered DOM. Rebuilding any of those from scratch would produce a worse version of something already excellent - the value is in the rule registry, the evidence model and the scoring around them.
A finding you cannot verify is a finding you have to take on faith, which is how audit tools lose developers. Each one carries the exact evidence that triggered it, the rule id and version, the date the rule was last reviewed against its source, and a link to that source.
A real browser
Your page rendered, its network captured and a screenshot taken
Core Web Vitals
Measured on the live page, not estimated from a formula
Two WCAG engines
One inside the browser against the rendered page, one over the markup
Structured data expansion
JSON-LD expanded to real schema.org types, not string-matched
RFC 9309 crawl parsing
robots.txt resolved the way a crawler resolves it
Durable score history
Scores kept, reports never - so a trend line costs almost no storage
Nothing here trains a model. Public datasets are used only to build regression fixtures that prove a rule classifies real-world edge cases correctly.
Founder
Computer engineer and full-stack developer · Pokhara, Nepal · working worldwide
I have spent seven years on the other side of this problem — building websites for clients, then running the ad budgets that send people to them. You learn something uncomfortable doing both: most of the money is lost before anyone clicks anything. A page that a crawler cannot index, a form a screen reader cannot complete, a link preview that renders blank. Nothing errors. Nobody is told. The traffic simply converts worse than it should, and everyone blames the campaign.
Every tool I reached for answered a slice of it, and each gave a number with no working shown. So I built the one I wanted: every point traceable to a rule you can read, every rule citing the standard behind it, and no language model anywhere near the score.
The part nobody else is covering yet is the newest one. Search is no longer the only way people find a business — an assistant answers first, and increasingly an agent acts. Whether those systems can reach your site, quote it correctly and complete a task on it is now a real commercial question, and almost nothing measures it. That is the gap this exists to close.
We run this on ourselves: the site is audited by its own scanner and the score is published rather than hidden. Two real bugs in the rule engine were found that way.
986 deterministic rules across eleven modules. Every finding comes with evidence, a citation and a fix preview.
Run a free scanNo signup. No credit card. Nothing stored but the result.