At the time, I failed to convince the original mapper to just tag the same wikidata=* value on each of the roadways. If a street needs a relation, I think street would be a fine name to call the relation type, but unfortunately someone got there first with a bizarre idea to put literally every node and way in a city inside one or more of these relations.
They were clients, not coworkers. Another Mapbox customer recently launched with “sidewalk differentiation technology”. I haven’t tried it, but I assume it’s also keeping sidewalks out of the guidance instructions. Like the client I supported, their application doesn’t display a map at all, so the user can’t tell that it’s actually calculating a route down the street. All they know is that it’s announcing street names instead of micromanaging their footsteps as if they’re a sidewalk delivery robot.
The Valhalla option on osm.org is configured with worse behavior: it sometimes traverses the sidewalk, sometimes traverses the street, and even the protected bike lane, rather unpredictably. Apparently the site requests routes with a completely neutral walkway_factor and sidewalk_factor of 1.0, making streets and service roads just as attractive as sidewalks. This gets you the worst of both worlds: overdetailed instructions and missing street names, plus extra turns that you shouldn’t make.
Experimentally, I find 0.25 to be a much more reasonable value for both factors if you actually want to traverse sidewalks, or a much higher value if you don’t. As someone who focuses so much on making things data-driven, I find that Valhalla’s geocoder-like emphasis on tunable factors with mysterious units really drives me batty sometimes, but it is quite flexible.