I finally got to understand the input paths of this scope better to check out my problem
On channel 2, there's a linear offset between the A and B channels of the ADC. It's around 9% of the full ADC range.
Zoomed in on the horizontal resolution, any time where we're plotting individual samples, I'm seeing a perfect sawtooth wave... When I change the vertical resolution, the sawtooth amplitude on screen matches. Positive and negative DC offset don't seem to change the peak-to-peak of the sawtooth.
This causes the first stage of the calibration to fail for channel 2 (but it succeeds on channel 1).
Considering that the A and B channels of the ADC are tied physically for the A&B in #A&#B in, and the VrefA&VrefB, I can only assume this skew is internal to the ADC chip

Is there anything in the firmwares that can correct for this offset? Or am I close to the end of my 1013d journey? Or have I misdiagnosed my issue?