Page requests have a per-address ceiling; programs use the API and the MCP server
Since 2 October the site admits page views per visitor address at a steady rate with room for bursts, and caps how many pages one address can have in flight at once. A browser never reaches these limits. A crawler that pulls thousands of pages in a few minutes now gets a 429 answer instead of slowing the site for everyone, as one did on the morning of 2 October. Programs that read our data should use the API with an account key, the datasets or the hosted MCP server, each with its own documented limits. Links also stopped preloading: a page no longer fetches every page behind a visible link, only the one you hover over or touch.