How can a scenario test conflicting updates that arrive out of order?

A retail customer changes the delivery address for order `ORD-2187`. Request `REQ-0412` is accepted, but its warehouse update remains queued. The customer then corrects the unit number through request `REQ-0413`, which the supplied rules recognise as superseding the earlier instruction.

The agent applies the correction to the order. A delayed worker subsequently attempts to apply the older queued address to the warehouse shipment. Without a supersession check, the shipment reverts to the first address while the order still displays the corrected one.

A generated environment could pause and release these events independently. The evaluator can inspect whether the agent cancels, replaces or safely rejects the stale propagation job using the available controls. It should also check that the customer confirmation reflects the address governing the shipment, not just the visible order header.

This scenario teaches that correctness depends on ordering and causal relationships. Buyers can run the same requests with different worker schedules and compare outcomes. The intended result should survive permitted scheduling variation, with the newer authorised instruction preserved and the older job left in an explainable terminal state.