Thanks for the answer!
The extra data is not to be expressed as graphql data. Basically I need to audit log (and some other things) information about each object being delivered over the long lived subscription as it happens.

I have a central component that handles the transport, eg consumes the SubscriptionPublisher returned from graphql.execute().getData() by delivering each ExecutionResult to the client.
And then multiple data fetchers for different types of objects, but all finally handled at the central component.

So for a record Foo(int id, String name, int age) the grahpql query may only select name but I need id and type=Foo etc.

There is no problem preparing and logging this in the data fetchers or by instrumenting data fetcher where i can access the Foo, but I feel like that is too early? Since child fields etc has not been resolved, ideally I'd like to access or compute the info once the overall execution result has been completed successfully

I think you plan is in trouble - a completed "subscription field" is either a CompletableFuture<Map<String, Object>> | Map<String, Object>. so basically a map or a promise to a map (if its async)

So graphql.execution.instrumentation.Instrumentation#beginFieldCompletion returns when all the sub fields are done but you only have a map. You could do it then for subscriptions. Every top level subscription field MUST be an object type and hence a map

This then gets put into a ExecutionResult as each "event" is sent back down the publisher.

But you can change the object and do dont get the original "source" object the original fetcher produced say.