The data that Neven showed on slack indicates that we have crosstalk in the PFS CCDs;  this is not surprising.  We should enable the correction, but of course this requires that we have the coefficients.  For now, I think we should use the HSC values (the PFS chips pretend to be 8-amp 4096*4096 chips, but are of course really 4-amp 2048x4096 chips, so we'd need to create a block-diagonal 8x8 matrix with the diagonal 4x4 blocks from HSC).

We should also measure them directly.  I think that the easiest way is to take some sparse slit very saturated arc data — saturated enough to have long bleed trails.  Assuming that we know the linearity well enough we can use these to measure the cross-talk coefficients.