Compute Schedule returns OperationConflict when two operations on the same VM are closed

We schedule one-time VM starts with Compute Schedule (virtualMachinesSubmitStart, api-version=2025-05-01).

If the VM already has a pending operation, a new virtualMachinesSubmitStart or virtualMachinesSubmitDeallocate with a deadline less than 60 minutes before or after it is rejected. The call returns HTTP 200, but results[0].errorCode is OperationConflict and results[0].operation is the existing operation. No new operation is created.

In our tests, 59 minutes apart conflicts and 60 minutes apart creates a new operation. It also conflicts across operation types (a start against a pending deallocate) and across callers (the VM's managed identity and a user).

We could not find these rules in the Scheduled Actions REST reference.

Steps to reproduce

1. Call virtualMachinesSubmitStart with deadline 12:00 UTC. A new operation is created.
2. Call it again on the same VM with deadline 12:30 UTC. The response has errorCode: OperationConflict and the operation from step 1.
3. Call it again with deadline 13:00 UTC. A new operation is created.

Questions

1. Is the 60-minute window by design? Does it apply to every Submit and Execute operation on the same VM, and is it documented?
2. How should we schedule a short power cycle, for example deallocate at 10:00 and start at 10:30?