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.
Why IP geolocation is a network estimate, not a street address
An IP address is assigned to a network, not to a person or a building. That single fact explains almost everything about what IP geolocation can and cannot do. When you look up an IP, you are asking where the network holding that address block is registered and operated, and inferring a location from that. You are not asking a device to report its GPS position, which is the only way to reliably get a street address.
Regional internet registries such as ARIN, RIPE NCC, APNIC, LACNIC, and AFRINIC allocate address blocks to internet service providers and large networks, not to individual homes. An ISP might hold a block that serves an entire metro area or region, and route addresses within it to customers dynamically as they connect. Looking up any one of those addresses correctly identifies the ISP and often the city, but it cannot know which building the connection currently terminates in, because that information never leaves the ISP's own systems.
This is why IP lookups return country, region, city, and organization rather than a house-level result. It is also why two devices on the same office network, sharing the same public IP, will resolve to the identical location, even though they might be in different rooms or different floors. The IP identifies the exit point to the internet, not the endpoint inside the building.
The practical accuracy pattern that follows from this is consistent across the industry. Country-level results are close to always correct, since address blocks are allocated within national boundaries. City-level results are usually correct but can be wrong when an ISP routes traffic through a regional hub in a different city. Anything more granular than that, a specific street or building from IP alone, should be treated as unreliable no matter which provider you use, because the underlying data simply does not support that resolution.
Good IP geolocation is honest about this ceiling rather than pretending to more precision than the data supports. My Geocode's IPv4 and IPv6 lookups return country, region, city, postal code, coordinates, time zone, ASN, and organization for exactly this reason, so you get the fields that a network-based lookup can actually support, at the granularity where it is reliable.
If your product needs a real street address, ask the user for one or request device GPS. Use IP geolocation for what it is good at: fraud signals, regional content, rough analytics, and defaults you can let a user override.