Scoped API keys are technically called scoped search api keys. So they are only used during searches and not during any other operations.

Yeah I am aware - What I am suggesting is that we could have

exclude_fields

as a scoped search feature for limiting what data is sent back using one of those keys - in this way the data is present in TS, but we can control the flow of whats returned. Seems reasonable. It won't improve performance for storing the data in the cluster in the first place if you really didn't want it taking up memory, but at least means you could keep the same collection with full data structure and allow different portions of it to be sent to search - rather than have to make duplicate collections with each subset of data.