The effect of sampling depth. Here we plot the residual deviance of the linear model applied to the error gradient of a 200 bp fragment for various step sizes. Each curve corresponds to a particular step size for a fixed sequence length. Notably, the effect of step size is independent of sampling depth, and further a sampling depth of 20 gives a good model for the reduction in error over sequence length for sequences in the order of magnitude considered in the benchmarks. Intuitively, larger sequence lengths will require greater sampling depths.