I would suggest to add some clarification on how the sentiment analysis algorithm processes the data.

For example, are the returned scores absolute meaning they are deterministic for the same backing model or are the scores relative to the other content/data in the given request. Meaning this particular document is more positive or negative than the other documents in the request.

Likewise for sentences of the document.

The reason this is important to me is that I'm running into the document limits (only allows 10) and wondering if I create a larger artificial document which is really combined data, then split it later, would this affect how it scores the sentences in the document.