About this tool

Purpose

This tool tests websites against the dotPublic Standards Framework. It produces a pass/fail verdict per check across eight categories: accessibility, accountability, AI & automation, interoperability, privacy, provenance, security, and transparency. Results are public so organisations can see how their sites compare and act on the recommendations.

How it works

Each scan runs three kinds of checks:

• Pattern checks — static matching of page content and request URLs against published tracker lists and regular expressions. No AI. Not interpretive.
• Behavioural checks — a headless browser visits the site and we record cookies, network requests, response headers, mixed content, and accessibility violations reported by axe-core. These are evaluated by rule, not AI.
• AI-assisted checks — for pages like the privacy policy, about page, or governance statement, the captured text is sent to Anthropic’s Claude model with a rubric. The model returns a pass/fail on each rubric item with a one-sentence reason, and for failures, a one-sentence recommendation.

AI use and human oversight

We use a large language model because some compliance questions (“is the privacy policy written in plain language?”) cannot be answered by pattern matching. Every AI-assisted result on this site is marked as such, and its reasoning is shown in full on the site’s detailed report, so you can judge the model’s verdict yourself.

We do not use AI to decide what checks exist, to rank sites, or to generate page content on this site. The check catalogue and the tool itself are open source at github.com/smagdali/compliance-tooling.