- 258 requests in 22 hours: what a crawl budget looks likeNovember 7, 2026
258 requests from Google's published IP ranges in 22 hours, against 12,362 English pages: 25.2% went to pagination pages and 15.5% to URLs that no longer exist.
- 98.5% of our "Googlebot" traffic was not GoogleNovember 7, 2026
6,121 requests claimed to be Googlebot in a 22-hour window; 90 came from Google's published IP ranges. The other 6,031 — 98.5% — were scanners hunting for .env files.
- 6,542 MCP servers, 62 install commands, 38 one-liners
- We ran 20 MCP install commands verbatim. Ten started.November 7, 2026
Twenty MCP install commands run verbatim: ten started, six failed, four could not be judged — and five of the six failures were introduced by the pipeline that republished the command.
- Seven stdio MCP servers, run verbatim: three answered initializeNovember 12, 2026
- We told Google not to index 10,538 of our 24,667 pages. Two batches of those were our mistake.
43% of our 24,667 pages carry a noindex tag. 307 were excluded by a rule that read a database column instead of the text a reader sees; 58 more had description text the template never printed.
- A ruler that cannot fail is not measuring anything
The two bugs that hid 365 pages were in the ruler, not the pages. Four negative controls, one empty input that must refuse to answer, and a rule that caught a change which moved zero pages.