Learned to run without a browser

The renderers on this site are client-only by design, which for a while meant a crawler saw an empty shell and a cold visit painted a skeleton.

Four processes, no special case

A doc-cache runner renders every document to HTML from a bounded pool of sockets. A seeder fills an empty site on first boot. An RPC runner answers calls from visitors who have no write access. A janitor patrols and repairs.

All four are ordinary peers holding ordinary documents — none of them is a privileged code path into the database. That is the part worth copying: if your server needs a back door into your own data model, the data model is wrong.