UFOPILOT

I had this problem with my M20J, it got worse over time, to the point the altitude delta was 200 feet. After much fruitless effort, it was finally found that the King HSI was sending bad pitch data to the KAP 150 autopilot, and the autopilot was dutifully chasing that bad data. I believe the 55A HSI has electromechanical brushes doing the sensing, and they can wear out, or otherwise go astray on you. This was solved by an overhaul of the HSI. Suggest you rule out any cheaper possible fixes first, as the King HSI overhaul is not cheap. I believe a solid state HSI such as the Aspen would eliminate this problem, by sending more reliable data to the autopilot than any electromechanical unit might be capable of. Compare the projected HSI overhaul cost against a solid state replacement unit.