Hi Thomas,
I could reproduce this issue. This is a but of an unlucky case where the fairly default low gap penalty is not sufficient to prevent a spurious alignment where the last part of the sequence get aligned in the wrong place to avoid mismatches with the reference. the reference happens to be clade A.D.2.2.1 and this pulls your sequence towards the reference. it doesn’t happen for longer or shorter sequences, because then the spurious alignment isn’t favorable.
I’ll increase the gap penalties in future version of the dataset to avoid these miscalls.
thanks for flagging this, this was instructive!
richard