OK, I see. About CUSUMSQ, this reference looks relevant:
https://doi.org/10.1017/S026646660808033X
(Deng & Perron 2008) It seems to have some max() formulations in it,
i.e. it seems to boil down the sequence to a single statistic.
But certainly not urgent after 18 years. Maybe we should create a
feature request.