Abstract We present a method for the in-flight relative flux self-calibration of a spectro-photometer instrument, general enough to be applied to any upcoming galaxy survey on satellite. The instrument response function, that accounts for a smooth continuous variation due to telescope optics, on top of a discontinuous effect due to the segmentation of the detector, is inferred with a χ 2 statistics. The method provides unbiased inference of the sources count rates and of the reconstructed relative response function, in the limit of high count rates. We simulate a simplified sequence of observations following a spatial random pattern and realistic distributions of sources and count rates, with the purpose of quantifying the relative importance of the number of sources and exposures for correctly reconstructing the instrument response. We present a validation of the method, with the definition of figures of merit to quantify the expected performance, in plausible scenarios.