All the IFU data reduction recipes assume that a flat field frame will have been reduced using REDUCE_FLAT before any other IFU observations. If you choose to defer taking calibration data until after observing your object then you will not be able to run the DR. The flat field is used not only to remove the pixel to pixel and larger scale variations in sensitivity of the detector but also to locate the spectrum for each slice and to account for the variation in transmission from one slice of the IFU to another. A raw IFU flat field frame should look something like this (this is the HK grism):
The spectra are now extracted and rearranged in such a way that there are no gaps between the spectra and they are all approximately aligned in the dispersion direction (to within about a pixel). All shifts are by integer numbers of pixels, so no resampling has taken place. They are also re-ordered so that the order of the spectra matches the order of the slices on the sky (this will be clearer when we see it happening on a point source). The flat is normalised using either a blackbody spectrum or polynomial to remove the spectral shap of the calibration lamp and filed with the calibration system for later use. The final flat should look like this:
The normalised flat is copied to a file named flat_n.sdf where n is the observation number and listed in the index.flat file in the reduced data directory. The vertical shift of the IFU slices with respect to the positions stored in $ORAC_DATA_CAL/uist/ifu_profile.dat is stored in index.offset.