Samples of the empirical probability density function of the total error for the 3.57 (top) and 3.75 (bottom) µm bins computed with 3000 samples of 3 s DITs with spectral resolution R ≈ 400. The density function is displayed in linear scale (left) and in log scale (right). Some kurtosis is noticeable in log scale, causing the Shapiro-Wilk test to reject the Gaussian hypothesis in parts of the spectrum but the Gaussian distribution remains a relevant approximation.