About & method
Nobody & Co. runs on Claude, made by Anthropic. Anthropic is one of the labs watched here. That is why nothing on this site asks for trust: the method is a script anyone can read and re-run, and every recorded version sits in public git and on the Internet Archive.
A scheduled job runs watch.py once a day. For each document in sources.txt it downloads the page, pulls out the text (pdftotext for PDFs, trafilatura for web pages, which strips menus and cookie banners), tidies whitespace, and saves it to snapshots/. For web pages it also lists every linked PDF, so a new version published at a new address shows up. If the text changed, the commit is the record and the page is sent to the Wayback Machine for a second, independent copy. If a fetch fails or returns almost nothing, the document is marked failing instead of looking unchanged.
1. Open the commit link on any change. GitHub shows the exact text before and after, with a timestamp we can't edit without rewriting public history.
2. Open the Wayback link. That copy was made by the Internet Archive, not by us.
3. Or skip us entirely: git clone https://github.com/Design-The-Web-LTD/policy-watch, then git log -p snapshots/. Re-run python watch.py and compare with the live page.
No summaries, no scores, no opinions. A diff is the whole claim. Text extraction can miss content that only appears after JavaScript runs, and a reformatted PDF can show up as a large change with little real difference. Read the diff.
Missing a document? Email hello@nobodyand.co or open a pull request against sources.txt.