Testing location data with edge cases, not just happy paths
An address in a well mapped city center tells you almost nothing about how your system handles a rural route, a disputed border, or a query near the poles. Test the hard cases deliberately.
Measuring coverage with a fixed set of reference addresses
Coverage sounds like it should reduce to a single number, some percentage of the world that a geocoder can successfully handle, but that framing hides more than it reveals. The only meaningful way to measure coverage is to ask a large number of specific, individual questions, can this exact address be found, at what precision, with what confidence, across a fixed and repeatable set of reference addresses, and to look at the pattern of answers rather than trying to collapse it into one figure.
Building a useful reference set starts with deciding what "coverage" needs to mean for your specific purposes, since a set built to test global consumer address lookup looks very different from one built to test coverage across, say, a specific set of countries you actually operate in. A reasonable reference set draws addresses from multiple countries, deliberately includes urban and rural examples in each one, and includes some addresses known to be difficult, apartment complexes, rural routes, newly built areas, alongside straightforward ones, since a set made entirely of easy cases will overstate real-world coverage.
The set needs to be fixed and reused over time, not regenerated for every test, because the value of this kind of measurement comes largely from comparability. Running the identical reference set against the same geocoder every few months lets you see whether coverage for a given region is genuinely improving, staying flat, or regressing, which a one-off test can never show you. It also lets you compare precision distributions over time for the same input, which is a far more informative signal than a single pass or fail count.
When you run the set, record more than a binary success or failure per address. Capture the returned precision and confidence for every result, not just whether an answer came back at all, since an address that resolves but only at city precision with low confidence is a meaningfully different outcome from one that resolves cleanly at house precision with high confidence, even though both would count as "found" in a naive pass or fail tally.
This is exactly the kind of testing worth doing yourself against any geocoding provider you are evaluating, including ours, using addresses genuinely representative of where your own users are, rather than relying on a vendor's general marketing claims about coverage. Check the coverage page for a general sense of what is supported, then validate against your own reference set through the forward geocoding endpoint before committing to any provider for a use case where coverage genuinely matters to your business.