Flux discontinuity in Kepler data

[emprice]

I've noticed that, frequently, the Kepler PDCSAP_FLUX has a few discontinuities (not gaps!) where the median flux level suddenly "jumps". This is currently causing detrending artifacts because a polynomial can't compensate for the step function. We're already throwing out any data with a quality flag set, so whatever event causes this isn't recorded by the telescope.

Is there a good way to handle this? Is there something in the instrument manual that could explain it/how to handle it correctly?