Hi,
AFAIU, the (UTF8 and ASCII) names of those administrative level 1/2 POR are the English translations when available, or the generic ones when the English translations are not available. However, when several English translations exist for the same administrative level 1/2 POR (see
http://www.geonames.org/2349961/anambra-state.html for a typical example), the database extraction algorithm seems to pick one randomly. As a consequence, from one day to another, the names of the administrative level 1/2 POR vary forth and back. With the given example, the names are chosen "randomly" from the {"
Anambara", "
Anambra", "
Anambra State", "
Udi State"} list. Which is... "less predictable" :)
May I suggest to just slightly amend the database extraction algorithm so that the join on the alternate name table be sorted and that the resulting extraction be more predictable? Note that the "p" ("preferred") flag could also be taken into account (be careful, though, that several translations may have the "p" flag set at the same time; so, the alphabetical sorting order may be a safe fall-back solution).
Thanks in advance
Denis