When we introduced the JSON format for a batch request we included a property, atomicityGroup, for grouping together requests within the batch that had to be executed atomically, and required that all such requests be adjacent. However that information is lost in the response; if the client wants to know which succeeded or failed as a group (i.e., they want to look for the request that caused a group of atomic requests to fail) they have to remember which requests were associated with which other requests. This is hard for a generic library to track and violates our principle that responses should be able to be fully interpreted without knowing the request that generated it.

The service has the information for the atomicity group, and must already handle the requests as a group, but the service currently has no way to convey this to the client. It makes sense to have the service return responses with the same atomicityGroup adjacent to one another, including the atomicityGroup identifier, to make it easier for the client to associate requests within a group.