Heavily masked profiles can trigger extraction errors

An extraction of sane flux in a heavily masked region generated an unmasked -6e9 pixel and thence a -90k scattered light trough. Fiber 1932, which was a good SCIENCE fiber. The fiberProfile is sane, covering 21 columns of which all but four are masked BAD. But the central pixel was not masked, so maybe minFracMask/minFracImage is not being tripped?:

row 174  cols 159..179
 image :  9  8 10 13 16 17 17 18 20 24 26 16  6  7  9 10 10 11 12 13 14   (all normal)
 masked:  1  0  1  1  1  1  1  1  1  1  0  1  0  1  1  1  1  1  1  1  0    -> 17/21 masked

This was to a test run with some non-final obs_pfs, so I'll just give coordinates: run=u/cloomis/1843-irp4_all, cam=n3, visits 144955 (irp4, shown) and 144959 (irp1) (same field). There is also a u/cloomis/1843-irp4_all_noInterp run, but that shows the same problem, since only the masked pixel values are different.

In the attachment, BAD pixels are pink. Kind of surprised this only happened twice (smaller one below), given the landscape.

It may also be that defect masks should fill in isolated good pixels.