A coordinator agent is repeatedly assigning the same subtask to multiple worker agents simultaneously, causing duplicate work. Traces confirm the coordinator receives each worker's completion signal but continues dispatching. What is the most appropriate remediation?

The traces confirm the coordinator is receiving completion signals — this rules out a communication or tool failure. The root cause is that the coordinator's instructions do not include logic to track task completion state before dispatching. The correct fix (option B) is to revise the coordinator's instructions to maintain a completion registry: mark each task as 'dispatched,' update it to 'complete' when the worker signals done, and check this registry before issuing new assignments. This is an instruction failure at the coordinator level.

Option A is incorrect because adding deduplication to workers is the wrong layer for this fix. Workers should be pure executors — adding orchestration-awareness to them creates tight coupling and distributed coordination complexity. Responsibility for task assignment tracking belongs with the coordinator.

Option C (replacing with a deterministic queue) is a valid architectural alternative but is an over-engineering response to what is fundamentally an instruction-level problem. A queue-based system has its own complexity and loses the flexibility of a reasoning coordinator. The question asks for the most appropriate remediation, which is instruction revision — not architectural replacement.

Option D (increasing dispatch delay) is incorrect because timing is not the issue. The traces confirm the coordinator receives completion signals in time; it just does not act on them correctly. Adding delay introduces unnecessary latency and does not fix the underlying state-tracking gap — after a longer delay, the coordinator will still dispatch duplicates.