What the campaign missed

⚠️ The result depends entirely on which building dataset you use

The first run of this analysis used OpenStreetMap buildings and produced a confident but wrong answer. OSM knows only 122 buildings in KANI district — yet the campaign recorded 15,683 visits there. You cannot visit 15,683 households in a district containing 122 buildings: OSM has simply never mapped these areas.

The consequence was a classic artefact — 64 of 67 "missed clusters" landed in the single well-mapped district, because that was the only place where a gap could even appear. The analysis was measuring OSM's mapping coverage, not the campaign's.

Switching to Google Open Buildings (satellite-derived) found 770,394 buildings vs OSM's 126,545 — 6.1×. The decisive validation: the share of visits that land nowhere near any building collapsed from 65% → 4.1%. Only then do the visits and the buildings agree, and the results below become trustworthy.

Coverage gaps by district

Share of buildings (in settlements ≥30 buildings) that sit in a settlement with zero recorded visits.

Largest unvisited settlements — field priority list

These are real, satellite-detected villages with no GPS-tagged visit within 150 m. Open the Map tab and click a red polygon for its exact location.

A second, independent problem: 23% of visits have no location

89,459 submissions carry no GPS at all — they can never be mapped or verified. The rate varies enormously by district, which points at a device/procedure issue rather than terrain:

• NIAKARAMADOUGOU 33%, MADINANI 31%, DABAKALA 29% — worst
• M'BENGUE 7% — best, and proof the other districts could do better

A handful of points are also plainly corrupt (latitude 50.8° — Europe; longitude +7.8°), and GPS accuracy ranges from 0.7 m to 8.3 km.

How to read this — and what it does not prove

• “Unvisited” means no GPS-tagged submission within ~150 m of the settlement.
• Because 23% of visits have no GPS, some flagged settlements may have been visited but recorded without a location. Treat the list as a priority list for verification, not as proof of failure.
• Settlements are buildings linked within 100 m; their outlines are convex hulls, so they slightly over-cover the real village.
• Buildings come from Google Open Buildings at confidence ≥ 0.65 (median 0.80).

Recommended next steps

• Verify the 258 clusters in the field — start with the 59 that have ≥100 buildings.
• Fix the missing-GPS problem: break the 23% down by agent and device to find the root cause.
• Agent anomaly checks: submissions/day, impossible travel speeds between consecutive points, and stacked identical coordinates (a fabrication signal).
• Cross-check against the Localité org units: a flagged cluster matching a known village with zero submissions is a very strong signal.