@LinusWeidickFrische @MapAmore
[Fair warning: I am speaking in an confident tone below; however, my accumulated knowledge of the data modeling philosophy behind OSM has largely occurred via osmosis, as I have slowly contributed and read more and more. I have not had formal, foundational education/training on the matter, so what follows might actually be somewhat scatterbrained.]
Presumably, the `_ref` bit was intended as an indicator of a "reference" value for the aforementioned `crossing`. E.g., `crossing_ref=zebra` where `crossing_ref` is the key, and `zebra` is the value, of the tag. In today's tag schema though, such tag keys are "flat" or "flattened," when they could instead be scoped/namespaced with colons: `crossing:markings=zebra` (presumably, one could have done `crossing:ref=zebra` as well, though `ref` has less meaning than `crossing`).
However, #osm eventually coalesced around a tree structure for their tag keys, with branches indicated by the colon (`:`) character and its subsequent string, and values for the last descendant/child node encoded as the tag's value. (I mean, the tags *might* be more general? Maybe there are cyclic graphs? But so far I haven't seen them (or at least, have not recognized them as such)). In theory, the underscore character (`_`) could have served the purpose of indicating branches as well, but... the English language uses spaces; tags cannot have spaces; underscores come closest to indicating spaces; ergo, use underscores as spaces and not as an indicator for a branch.
Incidentally, this approach also permitted:
1. namespacing of tags. E.g., one could namespace/scope specific tags so as to organize data. E.g., `us:tx:tarrant` could indicate the anything following (perhaps, `:parcel` to indicate a parcel number) relates to Tarrant county, in the state of Texas, in the United States.
2. having specific "branches" contain common semantic intent regardless of where in a tag's key tree they fall, such as `:height`. I.e., `building:height` or `antenna:height` or `building:antenna:height`.
Dive deeper into event sourcing with Ismael Celis on the latest Dead Code Podcast. 
A big thanks again to the BACH and eLaute projects 
We had two very fruitful discussions during our online meetings over the last weeks and could discuss some common problems in editing and data modeling of our individual projects. Looking forward to future exchanges! 
BACH: 
A big thank you to our team and all participants of our first workshop 'Modeling of Performance Data', which took place on November 8th at the University of Graz 
Take a look at our workshop report to discover who attended and which topics we discussed:
English Workshop report: 
you’ll actually want to get close to!
In zwei Tagen ist es soweit: Nach drei Jahren Projektlaufzeit präsentiert die interdisziplinäre "Digitale Kartenwerkstatt Altes Reich" ihre Ergebnisse. 
Hier geht es zum Programm: 
1/ The initiated already know what happens today, but do YOU? It's