Notes from building it.

Decisions and their reasoning, mostly. What was tried, what broke, and what the fix turned out to be.

Nothing published yet.

Rather than fill this with three invented articles, here is what is actually queued.

  1. Why per kilobyte, and what it beat

    Split evenly, then by time spent, then by data moved. The first two failed for reasons worth writing down.

  2. Every customer domain answered 522, for weeks

    Worker routes do not apply to Cloudflare for SaaS hostnames. One first-party subdomain hid it the whole time.

  3. The sign-in loop

    Four wrong hypotheses before the real one: signOut with scope local still reaches the server.

Until then, the developer docs carry most of the detail, and the working notes live in the repository.