Go to a bibliographic record and edit an item. In itemcallnumber type: a & Save. The result is, as expected, that the itemcallnumber will be literally 'a &' Edit the item again and change to a & b Save. Now the itemcallnumber will be 'a & b' as if the string has been subject to an extra unescaping of html-entities.
Not just itemcallnumber, it happens for every item field you can type in, and also for bib records. You don't need the 'b', just '& ' - without the space it isn't unescaped. And without the semicolon it isn't unescaped, though & is one of the things where browsers don't require it. And '& ' isn't unescaped, despite that also being defined. Nor are '< ' or '> '. I didn't try the whole list of named character references from the spec, but if < isn't, there's no reason to think that ϶ would be.