Which geocoded sites must be rematched before dispatch uses them?

The planning team has geocoded the Education Ministry's register of public fundamental-school buildings with Geocode Addresses in ArcGIS Pro 3.7 (Output Fields All), and the points are meant to replace the dispatch map's school layer. Acceptance is yours: hold the result to the team's rules below and flag every record that breaks one, with the rule it breaks. Flag, never fix — what you flag goes back to be rematched or to the register's owner.

Any GIS will do: ArcGIS Pro, QGIS or a script. You are graded on what you hand in.

The delivery

luxembourg-sites-geocoded holds one record per register entry, as points in WGS 84 (EPSG:4326). It carries a subset of the fields Pro 3.7 writes with Output Fields All. The match results in it are simulated for this exercise, in the shape Pro writes them; the register and its addresses are real. Every record is named by its id: whole numbers, one per record, no two the same.

The USER_ prefix is how Pro marks the fields of the table you geocoded; the others are what the locator returned. AddNum and USER_numero are written the same way (12, 3A), and so are Postal and USER_code_postal (1234), so each pair compares text against text, exactly as the file holds it. A record with Status U has no geometry.

luxembourg-communes is the country's communes as polygons, with commune (its name), canton and lau2 (its LAU code), in WGS 84. It is what the commune rule reads; it is not under review.

The rules

The locator's minimum match score was 85, so Pro matched only candidates scoring 85 or more. The team's acceptance rules are stricter:

The middle five rules read matched records only: a U or T record is flagged for its status and judged on nothing else. Test every record against every rule, and flag a record once for each rule it breaks.

Working it

• In ArcGIS Pro 3.7. Bring the GeoJSON in with JSON To Features, choosing Point as the geometry type; its first record carries every field, so the schema comes through whole. Records with no geometry may not come through as features, so compare the record count with the Features list here, which lists every record. Sort and filter the attribute table, or use Select Layer By Attribute — a field-to-field clause such as AddNum <> USER_numero compares text with text. For the commune rule, run Spatial Join with the sites as Target Features and luxembourg-communes as Join Features, keeping its defaults (Join one to one, Intersect), then compare commune with USER_commune. For shared ids, Find Identical on USER_id_source, or Summary Statistics with USER_id_source as the case field, whose Frequency field counts each id's records.
• In QGIS. Select by expression; Join attributes by location; count("USER_id_source", "USER_id_source") > 1 finds the shared ids.
• The downloads. The GeoJSON and the CSV keep every record and every field name. The Shapefile cuts field names to ten characters and leaves out the records with no geometry.

Choose a record on the map or under Features, read what it holds, and tick the rules it breaks; the list searches and filters, and flags every record it shows at once. A review worked out in your own GIS or a script goes under Paste a list, one id, rule a line, the rule by its id or its label. Each rule is graded on its own: every record that breaks it flagged for it, and none that meets it.