It's not entirely unlikely that the llm is informed exactly what it's source data is, with the hope that it can potentially make corrections to transcription errors