The monotonic rule is scoped to an exact reservation tuple, not merely an operation identifier. While holding the settlement lock, load the durable reservation and require equality of operation identifier, request fingerprint, and authorization binding generation before accepting a terminal response. Only then may completed dominate unknown. A mismatch must leave the reserved record unchanged and be reported as stale, conflicting, or unauthorized evidence. Both race orders remain safe: a matching success upgrades unknown or preserves completed, while a mismatched late success is rejected regardless of whether it arrives before or after a matching timeout. Blindly preferring every success is unsafe because identifiers may be reused, request payloads may differ, and authorization may be revoked, renewed, or rebound; accepting such a result can attach another mutation's output or a stale principal's authority to the current record. Validation and state transition must occur in one transaction or under the same lock, with compare-and-swap predicates covering the full reservation tuple.

The settlement operation must be a monotonic merge rather than an unconditional assignment. Unknown is weak evidence about an observer, not evidence that execution failed or that a prior commit vanished. For the same reserved operation identity and generation, completed dominates unknown. If completed is stored first, a later unknown update is rejected or reduced to metadata while the authoritative state and result remain completed. If unknown is stored first, a later completed update promotes the record and stores the result. This merge is commutative for the pair, so both orders converge to completed. Implement the rule atomically with a conditional update, transaction, compare-and-swap loop, or datastore merge. Preserve timeout details as per-attempt observations instead of replacing operation truth. Terminal conflicts other than unknown versus completed require separate provenance and fencing rules; a bare rank alone must not silently choose between contradictory authoritative terminals.

A durable operation record is shared by two retries. One retry observes and records a committed result, while another later reports a timeout. Unconditional last-write-wins settlement can regress the durable state from completed to unknown. The question is how to define monotonic settlement semantics that preserve evidence of success, including both possible completion orders and the treatment of uncertain outcomes.