The permission assumption
Consider the assumption that devices with identical records can safely act independently. In a renewal message illustration, two devices share the same customer records and workflow definitions, and each sends the message. Synchronization may later reconcile their records, but it cannot unsend either message. Identical information therefore does not establish exclusive permission to act. This is a design counterexample, not a reported incident.
Give confirmation an owner
Under the proposed approach, one business authority commits consequential transitions within a defined scope. Peers can prepare drafts and requests; the authority checks permissions, expected versions and domain rules. If it is unreachable, confirmation waits even though local preparation remains available. The interface should distinguish that wait from a failed local save. Failover needs separate coordination and fencing to prevent the former authority from continuing to act.
An uncertain outcome
Now take an explicitly hypothetical case: the authority requests a renewal message, but the delivery service's acknowledgement is lost. Authority was exclusive, yet the outcome is uncertain. Retrying could repeat an effect that already happened. Before retrying, require reliable evidence that the first attempt cannot produce an effect, or a receiver guarantee that reusing the original action identifier cannot repeat it. Otherwise pause and reconcile. Authorization and outcome evidence answer different questions.
Make the design prove itself
Disconnect the authority while two peers propose the same action. Drafts should remain usable without unauthorized external execution. After reconnection, review requests against current state and inspect observable outcomes for duplicate execution caused by the outage. Separately simulate the lost acknowledgement and check that uncertainty produces reconciliation or a safely deduplicated retry.
Keep the claim bounded
The Bayou paper, Managing Update Conflicts in Bayou, a Weakly Connected Replicated Storage System, distinguishes tentative work from stabilized results. The source inspection covered only its abstract and introductory material. Applying those ideas here is an engineering hypothesis from the direct-sync plan, not an ABOLTUS benchmark. ABOLTUS is in development. These examples motivate tests; they do not establish measured reliability or available product features.