Learn

What Is a Geocoding Confidence Score?

A geocoding confidence score is provider-specific metadata about how strongly a returned candidate matches a query; it is not a universal probability of correctness or a direct measure of positional accuracy.

Geobble
intermediateexplainerGeocoding Place Names & Identifiers

Explain how to interpret geocoding confidence, distinguish it from coordinate accuracy, and design review thresholds without assuming scores are comparable across providers.

What Is a Geocoding Confidence Score?

A geocoding confidence score is metadata produced by some geocoders to describe how strongly a returned candidate matches the submitted query under that provider's own ranking system.

It should not be interpreted as a universal probability that the result is correct, and it is not the same as positional accuracy.

What can influence a score

A provider may consider factors such as:

  • exactness of the address components;

  • spelling corrections;

  • missing or inferred components;

  • agreement between postcode, locality and region;

  • text relevance;

  • candidate prominence;

  • geographic proximity bias;

  • match type or feature class.

Different providers expose these ideas differently: one may return a numeric relevance score, another may use categories such as exact, high, medium or low, and another may expose component-level match codes instead of one scalar value.

The score is therefore meaningful only together with the provider's documentation and version.

Confidence is not coordinate precision

A query can match confidently while the returned coordinate is approximate.

For example, a provider may know exactly which street address you mean but only have an interpolated point along the street, making the text match strong while the location remains less precise than a rooftop point.

Conversely, a known feature coordinate may be geometrically precise while the text query is ambiguous and therefore receives lower match confidence.

Keep at least two concepts separate:

  • match confidence — how well the result matches the query;

  • positional accuracy/precision — how well the returned coordinate represents the real-world location.

Some providers expose both. Mapbox, for example, documents a match_code confidence for address-query matching and a separate coordinate accuracy classification such as rooftop, parcel, point or interpolated.

Scores are not comparable across providers

A value of 0.9 from one service does not necessarily mean the same thing as 0.9 from another.

The training data, indexes, ranking algorithms, calibration and candidate sets differ, and even within one provider, scoring semantics can change between API versions.

If you switch geocoders, revalidate your acceptance thresholds rather than carrying the old numeric cutoff forward mechanically.

Use thresholds as workflow decisions

A threshold should reflect the cost of a false match.

For a low-stakes exploratory map, you may accept a broad set of candidates and visibly flag uncertain records. For emergency dispatch, financial compliance or legal notification, however, an automatically accepted match may require much stronger evidence.

A practical workflow can separate results into:

  • auto-accept;

  • manual review;

  • reject/unresolved.

The thresholds should be tested against labelled examples from the actual geography and address patterns you expect.

Review more than the score

For any consequential result, inspect the evidence around the score:

  • returned feature type;

  • matched address components;

  • canonical name;

  • country/region hierarchy;

  • coordinate accuracy class;

  • alternative candidates;

  • original query;

  • distance from an expected area.

A high confidence result in the wrong country because the query lacked context is still wrong for your workflow.

Preserve the provider metadata

Do not store only the final coordinate.

Useful provenance fields include:

  • original query;

  • provider;

  • API or dataset version where known;

  • request date;

  • returned candidate ID;

  • match/confidence metadata;

  • positional-accuracy metadata;

  • reviewed/accepted status.

That lets you revisit results when the provider updates its index or your acceptance policy changes.

A score ranks evidence inside a system; it does not certify truth

Confidence metadata is useful because it helps triage large geocoding batches, but it becomes dangerous when a provider-specific ranking value turns into a generic “accuracy percentage”.

Treat the score as one piece of evidence. Combine it with component matches, geographic context and positional information, then decide whether the result is suitable for the consequence attached to it.

References

  1. Mapbox Geocoding API. Documents provider-specific address match_code confidence and separate coordinate accuracy classifications, illustrating why match confidence and positional accuracy are different concepts.

Related content