You only asked us how to tag “a pathway of unknown classification”, not whether highway=path represents such a pathway exclusively. In other words, this is “a pathway of unspecified classification”: maybe we don’t know the classification, or maybe we know it’s unclassifiable, whether because of a gap in our tagging scheme, or out a desire for fairness among all modes of transportation, or because the path is too informal to split hairs about.
To some extent, this parallels the unsolved problem of highway=service: is it a major service road that clearly doesn’t fit any service=* subcategory, or is it a service road we just haven’t been able to subclassify yet? There have been multiple attempts to define new service=* values, but they always seem to get hung up on a lack of words.
In the absence of new tags, the simple solution of choosing the best-fitting highway=*way tag may suffice in many cases. It’s like how we tell people to tag a hybrid store or mixed land use area as the thing that best describes its “general character” instead of comprehensively ranking everything that fits.
Yes, in hindsight, this ambiguity stems from the original highway=path proposal using the same tag for two different purposes at the same time, skunking it right out of the gate – or, seen another way, from using something as messy as access tags for iterative refinement of a primary tag. This was well before OSM data was ever used for automated routing on a global scale, back when navigation-related tagging discussions were dominated by a very geographically limited subset of the community. We’ve learned a lot since then, and a similar proposal probably wouldn’t fly anymore.
Many data consumers are designed for laypeople, so punitive mapper feedback could easily go unnoticed or result in frustration without any fixes. Shattering the pedestrian network outright isn’t really an option either. However, if a car routing profile does recognize highway=road, it should heavily penalize it, using it only as a last resort and showing a warning to the user. Likewise, a pedestrian routing profile should probably avoid a bare highway=path in favor of a nice highway=footway running parallel in close proximity. This is how some routing profiles already behave, so I’m unclear on whether there’s actually a problem.