OpenStreetMap
From the Build Plan developed with Claude
Project Purpose A web application enabling mappers and data quality analysts to select a geographic area, fetch Overture Maps POI data and OpenStreetMap data in parallel, algorithmically compare them, and produce two outputs: a reviewed, selective upload to OSM via OAuth2, and an annotated GeoJSON export classifying each Overture POI by assessment category.
No edits to OSM happen without manual review. Overture is more of an attention guide (at least that’s my hypothesis)
Lots of UX fine tuning to come. Quick observations:
- Overture is very noisy, lots that is irrelevant (business mailing addresses in personal houses, mislocated POIs, closed places, duplications with differences across Meta/Microsoft/4SQ sources)
- Matching is hard but can be improved
- There is useful signal in Overture, places that need addition to OSM in new developments, but you have to have means to easily sort through and keep a record.
I’m going to work on refining the workflow, add means to link features across Overture, ability to not just add but adjust OSM features, audit trail so it’s possible to share reviews and apply reviews against new releases….
Discussion
Comment from spiregrain on 7 March 2026 at 10:39
How confident are you that the Overture Maps data you are using are licensed in a way that allows you to copy them to OSM?
Comment from mikelmaron on 7 March 2026 at 11:35
Fairly confident but yes I would like discusssion https://docs.overturemaps.org/attribution/
Comment from SimonPoole on 7 March 2026 at 14:53
There’s already a fair body of discussion on the quality, or more the lack of it, of Overtures POI data.
The main problem are not the cases that can be reasonably be determined to be wrong with available additional information as you mention, but the ones that are plausible but are actually made up. Not all of them are so easy to verify as the nearby one, for which I can look out of the window and come to the conclusion that is isn’t a thing.
Comment from mikelmaron on 7 March 2026 at 21:40
Local knowledge is essential. You might counter, just go map it. For me, I suspect there’s value in the signal of what might be missing.
Comment from SimonPoole on 7 March 2026 at 23:36
The question is more how high is the false positive rate and what kind of level is acceptable.
I suspect nobody wants to waste significant amount of mapper time on chasing fata morganas.
Comment from RobJN on 7 March 2026 at 23:42
How about a website/app targeted at mobile phone users to allow them to quickly accept or reject Overture data while out surveying?
Comment from SimonPoole on 7 March 2026 at 23:58
How would you stop that essentially being doing Overtures cleanup for them?
Adding actually missing POIs is on the other hand trivial, just get the Overture data as GeoJSON and off you go, but as said potentially on a wild goose chase.