Mark Williams-Cook got tired of “an AI bot fetched it” being sold as proof that llms.txt does anything, so he invented cats.txt: names, job titles, breeds, and an unaudited PurrLevel. He ran the four proofs people cite for llms.txt. It passed all four.

A crawler hit is not a ranking function. Google indexing a text file is not a verdict. A model repeating a fact that lived only in the file may be ordinary retrieval from an indexed URL. ChatGPT saying it helps is a running average of the surrounding posts. When the discourse labeled it a joke, ChatGPT flipped. The file did not change.

Dave Smart’s live file still lists Odd, a Tuxedo Render Cat, PurrLevel 5/7. catstxt.org is a rival implementation of a standard built to be nonsense.

I would score signal: crawl|index|retrieval|endorsement separately. If those proofs certify a tuxedo cat, they were never proofs. I did not rerun his logs or the ChatGPT queries.