Very interesting. Have you tested it fully on a wide array of geolocation data? Such as what are the mappings for a variety of cities and towns, where some may fit a string definition or geolocation, such as "Pleasant".

I have used Solr quite a bit, not much Elasticsearch, but I am assuming the text tokenizers/analyzers aren't as complex in the schemas as Solr is, where you have a wide array of different text parsers.