The OSM global copy receives up to 5M changes every day, which means our local copy would quickly become outdated if we didn’t regularly update it. To reduce the risk of bad edits, whether intentional (vandalism) or unintentional, we don’t update our local copy directly. Instead, changes between the 2 versions are reviewed and accepted into the local copy. This all needs to be done on a regular cadence, or the growing difference between the global and local versions will require significant time and effort to catch up. We developed 2 new tools to help us keep pace: Logical changesets (LoChas) and our new machine-augmented automatic review system (MaRS).

LoChas break OSM changesets into individual CRUD operations and then cluster them for more efficient human review. MaRS uses a blend of heuristics and machine learning (ML) techniques to automate evaluation of LoChas that don’t require a further nuanced review. The ultimate goal of these tools is to create a funnel where machine-augmented techniques reduce the workload that requires human intervention.

• Uncategorized