A lot of new features are still not live. But this release fixes a couple of problems related to outliers. 1. In areas (primarily the ear canals) with very large Jacobians of the inverse field the predicted intensity could sometimes get "too large", which meant that "innocent" slices could get labelled as outliers. In some cases this would then feed back into the prediction maker, meaning that for each iteration that slice could get labelled in more volumes. Eventually leading to a situation where it was labelled an outlier in all volumes. I solved this by limiting the subtraction (observed-predicted) for the outlier detection to voxels where the Jacobian was less than three in the inverse field. The value is accessible on the command line as (--ol_jacut='some value'). 2. Related to the first problem, because eddy used pooled summary statistics for detecting outliers it would be "too sensitive" for low b-value volumes and "not sensitive enough" for high b-value volumes. In extreme cases, such as when very low b-value (~in the hundreds) shells with very few volumes (maybe ten or so) were acquired along with many (hundreds) of high b-value volumes, it could again lead to a slice being labelled an outlier in all volumes. I solved this by switching to shell-wise summary statistics for the outlier detection. This is now the default, but one can obtain the old behavior by specifying --ol_ss=pooled.