I dabble in geospatial data for my work in air quality modelling, accuracy for me means being within approximately 10-50 meters. Even under those very loose constraints I find the data handed to me by clients is often not good enough. Incomplete AutoCad dxfs, PDFs, and dubious SHPs all make the work much harder than it needs to be.
* Roads that have no physical delineations (e.g. an entire valley is the road).
* Landscape features can change rapidly (e.g. forest fires rendering roads unusable on a minute-by-minute basis).
* Landscape features can change slowly (e.g. erosion, continental drift)
* Landscape features can change slowly and rapidly (e.g. features on glaciers).
* Data that's flat out wrong for political reasons (e.g. disputed territories, GCJ-02, hidden "things").
* Data that's flat out wrong for technological reasons (e.g. GPS jamming).
* Things that are named in languages without written forms.
* Things that have official names other people consider offensive (e.g. Piestewa peak).
When you make a map, all of these and more become your problem. Every edge case anywhere in the world is potentially something you have to worry about.
This weekend I'm going to go explore one of the OSM tracks and see where it's a total hallucination, old and overgrown, or real.
I know many instances in which someone mapped a track through a field which subsequently got plowed by the field owner, but otherwise I have no idea. Could you tell any specifics?
I know just about enough GIS stuff to fit on a few index cards, but I quickly realized it is way more complex than how simple Google Maps makes it out to be.
With OSM I think a lot of walking tracks are from GPS traces uploaded by volunteers, so maybe sometimes stuff sneaks in. Or maybe the map shows old tracks that have since been erased by rangers. Or maybe there's a nearby track that has been marked down in the wrong place.