2015-09-03 7:05 GMT+01:00 Federico Leva (Nemo) <nemowiki@gmail.com>:
Jean-Frédéric, 03/09/2015 01:52:
Ok, here is the problem. The current design implies that one
configuration (country, project) has one row template and only one. As a
result, all monuments using WLM2014-riga and WLM2015-riga are *not*
parsed. This also explains the problems with categorisation: the bot
does find the monument in the database and thus cannot infer categories.

Looking at the WLM201X-riga templates, they appear to be perfect
supersets of one another − all the fields of 2013 are in 2014, which are
themselves in 2015. In this case, could we just unify the template?

I did not set up the templates, but at worst a wrapper template can be made.
However, can the bot fetch multiple sources (multiple roots for the lists) and merge data from multiple rows for a single ID?

No. I believe the latest crawl would replace previous values for the same ID.

In the meantime, I have changed the config to harvest the WLM2015-riga template. This has added some 3000 more monuments to the database [1] for Italy.

The bot also categorized hundreds of pictures from Italy, but I stopped it because of concerns raised on my talk page. See column 'Bugs' in [2]

[1] https://commons.wikimedia.org/w/index.php?title=Commons:Monuments_database/Statistics&diff=prev&oldid=170596671
[2] https://phabricator.wikimedia.org/tag/wiki-loves-monuments-database/