A request filed by a script or an agent now runs

Several things in the platform are asked for by writing the request down: you set "activate this subscription", "enrol this person", "cancel this job" on a node, and the part of the platform that owns that node notices and carries it out. That is the normal way work gets requested here, and it is deliberately the same whether the request comes from a page, an agent, a script, the API, or a billing webhook.

It only actually ran when the request came from a page.

The piece that watches for these requests starts when the node it belongs to is first opened. A page opens the node as a side effect of showing it, so the watcher was already running and the request was picked up within seconds. A request created any other way wrote the node correctly — and then nothing opened it, so nothing was watching, and the request sat there. Not failed, not queued: sat, with no error, no failed status, and nothing in any log that would have said so. It would then run, correctly and immediately, the next time anybody happened to look at that node — which could be minutes or hours later, or never.

One measured case: a subscription activated by an agent sat untouched for four and a half hours, then completed twenty seconds after the first person opened it. The same subscription created from the admin page had activated in thirty-one seconds. The only difference was who asked.

The people this hit are exactly the ones least able to see it: an operator running a script, an agent doing a task on your behalf, and an automated payment confirmation. The visible symptom was somebody who had been granted a plan and still saw the paywall.

A node created with a pending request now wakes its owner, so the watcher sees the request and acts on it straight away regardless of who filed it. Two details worth knowing:

Reconnecting…
The server was updated. Reloading the page to pick up the latest version.