Use of this option forces tckstats to explicitly calculate the length of every streamline based on the distance between every sequential vertex pair, rather than relying on track file header information regarding step size to speed up the calculation. This is particularly important if some non-rigid transformation has been applied to the track data.

In response to this forum post. Ended up opting against wiping the track file header fields in tcknormalise: The tracking algorithm step size is still relevant information despite a non-linear transformation having been applied, so even if output_step_size were removed (which makes sense), tckstats would then revert to using step_size to calculate streamline length.

There are of course other ways this issue could be addressed (e.g. never use step size information when calculating length); but all have their pros and cons (e.g. calculation of histogram bin size works best if the step size is known). Feel free to comment if you think one of the alternatives is preferable to this.