Acknowledgments
We thank CJ Perovich for thoughtful discussion, and IA Sutton for help with maximum likelihood methods. This work was funded by an NSF EAGER award (DEB #1837233) to JSW.