The "Try again" action on failed generation tiles is now fully functional. Until today, clicking it surfaced a "someone already retried this" message even on jobs that had never been retried, because the retry pipeline could not actually move the job back into the pending state. The fix routes the retry through the correct privileges and now reserves credits, requeues the job for the worker, and lets the tile transition back to in-progress as expected.
If you have older failed generations in your product hub, you can now retry them directly — no need to delete and recreate the order.