Sample distribution in empty windows affected by gate 5 activation (and hence limited in integration time to 32 ms) when they are bias corrected by scalar prescan level only (top) and when they are bias corrected using the full bias instability model (bottom). The sample distribution in the latter has a near-normal distribution (disregarding the negligible few outliers on the positive side of the distribution resulting from stray photoelectric flux and prompt-particle events, etc.) with Gaussian equivalent σ = 5.2e⁻ and is dominated by the video chain read-noise-limited performance for that device (see Hambly et al. 2018 for further details). In both panels the magenta line shows the cumulative distribution.