Coordinate precision and floating-point rounding errors
Storing or computing coordinates using the wrong numeric type can introduce rounding errors on the order of meters, entirely separate from any error in the original geocoding.
Storing or computing coordinates using the wrong numeric type can introduce rounding errors on the order of meters, entirely separate from any error in the original geocoding.
Rooftop geocoding places a point at the actual building. Street-level geocoding interpolates a position along the road. The two look similar on a map and are not the same thing.
Calling a result high precision means something specific. It should mean the match resolved to an exact building or point, not merely that a response came back quickly.
A confidence score is not a single measurement, it is a combination of several signals about how sure a match is. Understanding those inputs makes the number far more useful.
A rural address, a landmark, or a new development without an assigned house number still geocodes correctly, just at a coarser precision than a full street address.
A response with fifteen decimal places of latitude looks precise. It is not accurate just because it has more digits than the method can actually support.
A pair of coordinates looks like the whole answer, but without knowing how they were derived, two identical-looking coordinate pairs can carry very different reliability.
Trimming decimal places on a coordinate looks harmless, but each digit represents real distance on the ground, and rounding too aggressively can move a point by meters or kilometers.
Precision tells you what kind of match you got. Confidence tells you how sure the geocoder is. Mixing up the two leads to bad filtering logic.
Accuracy claims mean nothing without a method. Here is how reference addresses with known coordinates are used to check what a geocoder actually returns.