We derive the full covariance matrix equations for proper treatment of correlations in signal fitting procedures, extending the results from previous publications. The straight line fits performed with these matrices demonstrate that a significantly higher signal to noise is obtained when the fluence exceeds , in particular in long (several hundreds of seconds) spectroscopic exposures. The improvement arising from the covariance matrix is particularly significant for the initial intercept of the fit at , a quantity which provides a useful redundancy to cross check the signal quality. We demonstrate that the mode that maximizes the signal-to-noise ratio in all ranges of fluxes studied is the one that uses all the frames sampled during the exposure. While there is no restriction on the organization of frames within groups for fluences lower than , the co-adding of frames should be avoided whenever the fluence exceeds this value.