Found the error!
Turns out that dataset 0.35m was broken.
And get pretty much a perfect fit without it at low orders.
Here’s the dirty scan:
it should look something like this:
It seems to have missed a data frame. This should have come up in the error checking…