Extended Data Fig. 2: Goodness of fit computation for the best-fitting baseline + Sgr dSph model.

These use our preferred set of templates (first entry in Table 1). In each of the 15 panels, one for each of the energy bins in our analysis pipeline, the blue histograms show the distribution of - ln L values produced in 100 Monte Carlo trials where we use our pipeline to fit a mock data set produced by drawing photons from the same set of templates used in the fit; orange dashed vertical lines show the 68% confidence range of this distribution, and black dashed vertical lines show the mean. Under the hypothesis that our best-fitting model for the real Fermi observations is a true representation of the data, and that disagreements between the model and the data are solely the result of photon counting statistics, the log-likelihood values for our best-fitting model should be drawn from the distributions shown by the blue histograms. For comparison, the red vertical line shows the actual measured log likelihoods for our best fit. The fact that these measured values are well within the range spanned by the Monte Carlo trials indicates that we cannot rule out this hypothesis, indicating that our model is as good a fit to the data as could be expected given the finite number of photons that Fermi has observed.