I found that replacing geometries by their string "signatures" is very efficient technique when comparing geometries. E.g. Assign point IDs to respective start and end attributes of a polyline or finding points that overlap. This approach combined with dictionaries is a game changer.

By some reason, that I don't fully understand, truncate produce more robust results than rounding or converting decimal to integers described in my comments. Thus this:

def truncate(f, n):
 s = '{}'.format(f)
 i, p, d = s.partition('.')
 return '.'.join([i, (d+'0'*n)[:n]])

'============================================

truncate( !Shape!.centroid.X,2)+truncate( !Shape!.centroid.Y,2)+truncate( !Shape!.area,2)

will potentially work better, when populating field for dissolve. In fact it is enough to summarise it and find minimum of sequential integer value stored in other field (FID and OBJECTID won't work when using standard Summarise tool, but are perfectly fine in script!). This way first unique features can be found and exported for further analysis

I found that replacing geometries by their string "signatures" is very efficient technique when comparing geometries. E.g. Assign point IDs to respective start and end attributes of a polyline or finding points that overlap. This approach combined with dictionaries is a game changer.

By some reason, that I don't fully understand, truncate produce more robust results than rounding or converting decimal to integers described in my comments. Thus this:

def truncate(f, n):
 s = '{}'.format(f)
 i, p, d = s.partition('.')
 return '.'.join([i, (d+'0'*n)[:n]])

'============================================

truncate( !Shape!.centroid.X,2)+truncate( !Shape!.centroid.Y,2)+truncate( !Shape!.area,2)

will potentially work better, when populating field for dissolve. In fact it is enough to summarise it and find minimum of sequential integer value stored in other field (FID and OBJECTID won't work when using standard Summarise tool, but are perfectly fine in script!). This way first unique features can be found and exported for further analysis