A delivery our endpoint did not answer in time was marked failed after 3 attempts within about 60 seconds, and was never retried after that. An endpoint that is briefly down or slow to start then loses the event for good, unless someone notices and resends it by hand.
Please retry for longer, with exponential backoff over hours (as most payment providers do), or let us configure the schedule. Until then, could the docs state the current retry schedule exactly (number of attempts and the delays between them)?