I... don't have this issue. I do see your example is a substat. Are you outputting the parent stat object?

Why aren't you filtering out duplicates in your program? I'm just doing something simple like:

Code
    exists = []
    for obj in new_stat_objects:
        value = obj.find('fields/field[@name="Name"]').get('value')

        if value not in exists:
            exists.append(value)
        else:
            # find duplicate fields, then remove the stat_object parent with the least number of fields
            # the assumption is that more fields = the most up-to-date stat_object, so we keep that one
            dupes = new_stat_objects.findall('stat_object/fields/field[@name="Name"][@value={}]'.format('"' + value + '"'))
            if len(dupes) > 1:
                field_counts = []
                for dupe in dupes:
                    parent_obj = dupe.getparent()
                    field_counts.append(len(parent_obj))
                i, v = min(enumerate(field_counts), key=operator.itemgetter(1))
                new_stat_objects.remove(dupes[i].getparent().getparent())


Also, are you reindexing the stat objects? (My script processes multiple .stat files and merges same-name .stat files together, so I have to reindex.)

Code
def reindex_output(file_path):
    try:
        e = etree.parse(os.path.abspath(file_path), XML_PARSER)
    except etree.XMLSyntaxError:
        raise etree.XMLSyntaxError

    root = e.getroot()

    i = 0
    for stat_object in root.iterfind('stat_objects/stat_object'):
        stat_object.set('index', str(i))
        i += 1

    with open(os.path.abspath(file_path), 'w', encoding='utf-8') as f:
        f.write(etree.tostring(root, encoding='unicode', pretty_print=True))

LSLib Contributor | My Mods: DOS2, DOS2DE | 560K Steam Workshop Subscribers