Waveforms and Filtering

I'm trying to measure features of the waveforms for clusters sorted by kilosort2/phy, and I've noticed that the waveforms returned by model.get_waveforms look different from spikes obtained by windowing the raw data at the spike time.

I can see the subtraction of the window median, in the get_waveforms function, and I can tell that there is some removal of noise that appears across the electrodes, both of which I want. There are also some subtle changes in the mean waveform shape that I can't find the reason for, but look like maybe some kind of filtering. I can also see that phy allows you to show waveforms with and without high pass filtering, but I can't find that option in get_waveforms.

Is there a full list of the processins steps between the raw data and the waveforms shown in phy and those returned by get_waveforms? Is model.get_waveforms the best way to extract all waveforms, with the goal of extracting waveform properties from the average waveform?