In my two instances it appears my PP was understated by several hundred watts at the time.
That straightened itself out after a couple BT workouts.
90 days or more of historical data is best, but that data needs to include sufficient maximal efforts under fatigue for the signature algorithm to work best. The alternative is to seed your data with those efforts now.
See this related post – Let's See Some Recent BT's - #252 by ridgerider2