The hypothesis is that evidence-aware candidate triage will increase useful, independently adjudicated OSM corrections per reviewer-hour while reducing unsafe false positives. This is unproven.
- Results to date: 312 directional hypotheses; 172 distance-only hypotheses; 60 blinded review locations; 0 independently verified errors; 0 accepted changesets.
- Baseline: elementary endpoint-to-endpoint proximity. It is not Osmose, MapRoulette, or another mature competitor.
- Satellite imagery predates the OSM edits on two inspected cases; a visible dirt-looking corridor is not proof of vehicle access. Those cases remain unverified.
- Two independent qualified mappers and an independent adjudicator are required. Missing labels are not negative labels; precision stays
null. - Human/community validation before any OSM task publication. Do not publish sensitive real-site rankings, patient locations or imagery without license.
- There is no evidence so far that emergency services or humanitarian responders benefit from any modeled connection.
Next stage: publish complete offline test harness, request HOT/OSM feedback, perform a small approved blinded review, and decide stop/pivot/scale on accepted useful corrections per reviewer-hour.