I have a custom API provider that contains not only the output of the prompt but also relevant context such as the data used to answer the prompt, the query used to obtain the data, and examples used to generate the query. I want to be able to use those pieces of context in promptfoo. For example, I might want to use the context-faithfulness metric to test if the generated output used the provided data.
It doesn't look like I can use the output dataset context from my custom provider in tests. Is there a way to use additional fields returned from a custom provider in tests? One way around this might be to save the data out to a CSV each run and to reference that CSV file in the tests but I'd rather avoid doing that.
Any ideas how I can use these additional pieces of information in my tests?