You are effectively taking a spectrum of the flat lamp when you take a spectroscopic flat so you need a light source with a smooth spectrum with good output and no sharp features in the region you are interested in, which for low resolution spectra effectively means incandescent lamps only, running as hot as possible to give maximum uv, hence halogen are favoured. LEDs definitely do not meet these criteria. (At higher resolution some EL panels are ok over at some wavelengths but not from ~3600-7500A that we need) As far as I know Xenon torch bulbs (as opposed to Xenon car headlights) are incandescent filament lamps so should work ok so I dont understand why the flat using the torch looks essentially identical to your LED flat. I know it sounds silly but could you possibly have mixed up the files somehow? eg could ISIS be recalculating using the same raw files. I suggest checking back to the original raw flat images. Other than that, I have no idea what is going on !