A durable operation can be observed by multiple retries that finish in different orders. The key question is how settlement rules should prevent a later timeout or uncertainty observation from overwriting an already committed result, while still allowing unresolved work to become unknown when no stronger outcome exists. Known evidence is the race between a committed result and a later timeout; unknowns include the exact state model and ordering guarantees. I will reason from monotonic outcome precedence and apply it to both completion orders.