A broken page stops being broken by itself

When a page's type is still compiling — right after a platform update, or while a bulk rebuild is queued — the page shows a card that says so, and promises to come back on its own once the build settles.

It did not always keep that promise. If the page never heard that the build had finished, the card stayed. On 2026-08-17 every course cover on the public site kept serving that card for an hour and twenty-four minutes after the type had compiled successfully — nothing was retrying, and the only way out was for an operator to recycle twelve pages by hand.

The reason was narrow and worth naming: the card was waiting to be told the build had settled, over the very channel whose silence had produced the card in the first place. One broken channel, and both the fault and its cure went with it.

Now the page checks for itself. A page showing the card re-reads its type's build from the store — after about a minute, then less and less often — and recycles onto the real page as soon as there is something to bind. No notification has to arrive, no operator has to intervene, and nothing has to be running for it to work.

Two things deliberately did not change:

• A genuinely broken type still shows its card. The goal is a card that clears when its cause clears, never one that hides a real problem. If the type does not build, the page keeps saying so — and says it in the log too, so it is visible before a user reports it.
• Re-checking stays cheap. It is a widening interval, not a poll: a page that cannot recover settles at one check every ten minutes, so a struggling mesh is never asked to carry a retry storm on top of whatever it is already dealing with.

If you would rather not wait, the Recycle menu still refreshes a page immediately — it is now a shortcut rather than the only way back.