Identity has a complicated relationship with both form and function. When I was young, my parents would stick a label on various things around the house to help me learn their names in English: “bookshelf”, “piano”, “piggybank”, “cookie jar”. Ironically, the bookshelf held as many picture frames as books, but the store sold it to us as a bookshelf. The piano was technically an electric keyboard, but I practiced piano on it. Unlike the stereotypical piggybank, mine looked like a postbox, painted green to represent the dollar bills that went in. It did not oink like a pig. However, the battery-powered cookie jar, resembling a barn, did make cute animal noises whenever I removed the lid.
Our wiki serves mappers as sort of a field guide: if you see something like that, then use this tag. But people sometimes take this guidance too literally, leading to misguided attempts to define feature types more scientifically. Once, I overheard the wiki described as a guide for aliens who had just arrived on Earth. Aliens probably would appreciate a definition of shop=bakery that specifies the maximum sugar content of the baked goods, to avoid confusion with shop=pastry, or a definition of shop=* that requires customers to come away with shopping bags, to avoid confusion with amenity=*. These definitions are useful as rules of thumb, but they become harmful when people interpret them pedantically. Trucks in a shopping bag?
Lest anyone thinks I’m exaggerating, tactile_paving=* currently has a definition that is unusable in the field and unworkable in countries that aren’t named France. Worse, fuel:octane_*=yes used to be universally defined in such a way that, in some countries, mappers would have had to engage in criminal corporate espionage instead of simply reading the label on the fuel pump. Whenever the wiki oversteps the bounds of reason due to a lack of global perspective, we can safely assume that mappers simply ignore the wiki.
There’s no such thing as a universal, language-agnostic dictionary, so there’s no point in turning the wiki into one. All we can do is offer enough hints that the mapper will successfully relate them to something they observe based on their previous lived experiences.
This is a good rationalization of the building=* scheme, though I don’t think most mappers think about it all that much. As long as the preset is called Church Building, most mappers will use it for anything they intuitively consider to be a church building. Maybe they distinguish between as-built and converted church buildings, maybe not.
One of the reasons mappers care about building=church, building=train_station, etc. is that some renderers, especially OSM Carto, give them a more prominent color as “significant” buildings. OSM Carto introduced this treatment as a compromise when deemphasizing buildings overall. It uses these values as a proxy for important landmark buildings, applying a traditional European perspective globally.
At the time, I did a rough back-of-the-napkin calculation to find that only a quarter of religious buildings in Cincinnati, a relatively religious U.S. city, could be considered historic landmarks, making building=church a worse proxy for importance than building=school. (Traditionally, American topographic maps mark places of worship, schools, libraries, post offices, town halls, and fire stations as significant landmarks. As a result, these facilities are considered “community anchor institutions” for various purposes.)
My takeaway is that the values are nearly useless to global renderers that aim for an unbiased presentation. Still, they would still be useful for helping a geocoder label e.g. “Church Building” in search results, leaving the specifics to the user’s imagination.