I recently discovered the mapbox feedback page and saw that they require you to classify the nature of your feedback. One click places the feedback on the map, and another click adds an instruction with your comment: * add a place (create a new feature with a list of tags) * fix a location (overwrite a feature with a list of tags) * rename a place (overwrite a feature with a name value) * fix satellite map (add a comment for satellite mappers) * remove a place (delete one feature with name value) * it’s something else (add a comment to OSM or something)

If users categorize their comments on OSM by type of instruction, AI could more easily interpret a comment string and learn to apply fixes by correlating human edits with comments.