Webhook delivery used to be a single shot. If your endpoint timed out or returned 5xx, the event was marked failed and Probo moved on.
Deliveries are now durable jobs. Each subscription gets its own delivery ID, Probo keeps that ID across retries, and it sends Idempotency-Key plus X-Probo-Webhook-Delivery-Id so receivers can ignore duplicates. Transient failures — network errors, timeouts, 408, 425, 429, and 5xx — retry with exponential backoff, up to 12 attempts. Retry-After is honored on retryable responses.
A successful delivery is any 2xx within 15 seconds. Other 4xx responses still fail immediately. Status stays PENDING while retries remain, then FAILED when the job is done retrying.
If you already keyed idempotency on eventId, you do not need to change that. If you treated every HTTP POST as a new event, start using the delivery ID before creating side effects.
