Fewer records, more of them real.

The brief sets the boundaries; everything downstream inherits them. Location fields are made consistent before any comparison runs. Format is tested on a sample before the full run is delivered.

Known-gap register

A written list of what the file does not contain, kept with the file.

Address normalisation

Standardising location data before comparison, so one place does not read as several.

Match confidence

Separating confident pairs from probable ones, and reporting the latter rather than folding them in silently.

Start with a single segment and judge the rest on it.