Since you support merging fields you likely would want to track provenance (including timestamp) on a per-field basis. Perhaps via an ID for the originating request.

Although I would suggest that rather than merge (and discard) on initial lookup it might be better to remember each individual request. That way when you inevitably decide to fix or improve things later you could also regenerate all the existing records. If the excess data becomes an issue you can always throw it out later.

I say all this because I've been frustrated by the quantity of subtle inaccuracies encountered when looking things up with these services in the past. Depending on the work sometimes the entries feel less like authoritative records and more like best effort educated guesses.

I’ll definitely discuss this with Drew, as he’s the one working on the database refactor. Thank you for the feedback!