Sky subtraction doesn't take fibre sensitivity into account

The 1-D sky subtraction code doesn't seem to normalise the spectra correctly. I took one of the September PFI visits (68879) and labelled all the fibres in the third to tenth percentile of flux as SKY, then reduced the data.  The raw data (pfsMerged.flux with no SKY fibres) are   and the sky models (pfsMerged.sky) are  .  The sky-subtracted spectra (also pfsMerged.flux) are   and the problem is clearly seen.