@Marily Nika First I would look at our eval set to understand what is wrong and why, and asked myself/team does it change our platforms intended goal? Or does it still meet it? If it does ship it with that understanding while you continue to monitor the outputs to collect more better on how to make it better. If it doesn’t then look further look at those use cases to understand if the app is designed correctly to achieve more consistent results or if they or the app needs to change based on it. But that gets could get messy quickly.