Latency in reset frames for some flight settings

In some ramp sequences, areas that received moderate levels of flux would reset to a background value *higher* than the usual background of ~11000 ADU.

Here's one example:

gl2591.flat.40101.fits

This is a flat field at a hi-med setting at 1636cm-1, taken with the pattern "N6 S8 N5 D0".

On blackbody, the first frame of the ramp looks normal, with the background level around 11000, and some small accumulation of bb flux subtracting from that value.
On a bright part of the blaze, median(image(410:450,600:700))= 10860 ADU

At the start of the ramp: median(subim)=10860

At the end of the 1st ramp, median(subim)=8710

At the end of the ramp, frame #14, the counts in the box = 8710 ADU

After reset following ramp #1, median(subim)=11164

And then on reset, frame#15, the median counts in the box are=  11164 ADU