Originally Posted by MCTuomey
... I underexpose (-2/3 ex comp) my Q files, then pull up the shadows if needed in post, as you described.
This effectively reduces raw data signal-to-noise ratio and dynamic range by at least 2/3 EV compared to the highest practical exposure.
Banding is typically more obvious as SNR decreases. In all cases banding appears first in shadow regions because they have the lowest SNR.
Think about a raw file made with the lens cap on. There is no light, so there is no signal. In fact the time dependent (random) read noise and time independent artifacts (banding) become the signal! Now both could be well above the analog-to-digital converter's noise floor, both are digitized and both appear in the rendered image.
With the lens cap removed the light becomes the signal. At some level of illuminance, the signal and the photon noise levels become much greater than the read noise and artifact (banding) levels. The artifact levels fall below the ADC noise floor threshold and they are not digitized.
Here's a link
Banding that increases with exposure is periodic noise. Periodic noise is caused by interference between coherent electronic artifacts and the signal. In some cases interference could increase with signal level (exposure). Periodic noise can be minimized by in-camera data filtering.