Or an early Norman one, for the same reasons. The people in the fort were different to those outside, city walls were built later in the medieval period once those differences had reduced.
I did try asking one of the PIs whether they were doing it to demonstrate that it doesn't work in order to get paper authors to provide their real data but didn't get a reply.
Will be going to a conference at the end of the month where there will be several presentations on the use of LLMs for this.