The flux level of my flux-calibrated data differs substantially if I use daytime flat fields instead of the attached night time flat fields. Why?

This effect is know for volume-phase holographic (VPH) grisms and is caused by different flux levels in ADU/sec between daytime flat fields and attached night time flat fields.

The response of VPH grisms depends on the position of the target on the detector along dispersion axis. Because science data and flux standard stars are observed at different positions this effect needs to be corrected during the flux calibration, because otherwise the shape of the flux calibrated data may be wrong. This is done by scaling the science and standard star data with the spectral energy distribution (SED) of their flat fields. In order to keep the overall flux scale the flat field SED is normalized by its flux at a given reference wavelength, taking into account the slit width and exposure time. The normalization factor applied to standard star and science flat fields should be similar provided that the lamp flux in ADU/sec is similar.
However, the flux in ADU/sec varies significantly between attached flat fields and daytime flat fields, thereby distorting up the flux scale.