Today there's a half baked calibration data cache which is in the wrong place. Move the caching to the persistence layer.
There should also be a mechanism for detecting if the calibration data has been changed "under our feet" by someone directly modifying the database and trigger a reload.