This was a 2-part bug: 1) "Track index" (i.e. "index of NLA track within NLA Stack") was calculated incorrectly when the transform data is created. A typo meant that this was actually "index of strip within original track" 2) Strips were flying around madly vertically (especially for downward movements, and with extreme difficulty for upwards) as the track index was being updated incorrectly. I suspect at the time I was misreading it as trackIndex = ... vs trackIndex += ..., though it obviously looked wrong looking at it now!