I'm automating a sales proposal software that takes the transcript from appointments with sales reps, extracts structured data and matches tasks based on a pricing index of over 40 actions for accountants. This then goes into Airtable for review from a senior with quotes from the transcript as reasoning. My issue is I need to get my extraction process to be more accurate, and I have written a script that automates pulling the proposal from ignition and the script and pairs them together but in some examples because of discounts for loyalty or other reasons it means that the harness cannot be thorough enough. I am rather new to this, so how do people usually setup harnesses to verify accuracy and see where improvements are needed. Thanks.