Interpret a decision

crosswalk() returns candidate edges with two labels.

Label	Meaning	Appropriate action
match	The candidate cleared the configured score and evidence gates.	Apply only under a policy that accepts the measured risk.
review	The candidate is plausible but lacks sufficient evidence for an automatic link.	Send to an accountable review process.

Pairs that fall below the review floor are not included in the result. They are not labelled different, because the resolver did not make a negative identity claim about them.

Score is not a probability

Treat score as an internal ranking signal. It is useful for prioritising review, but it is not a statement that two records have a particular probability of being the same entity.

Inspect the evidence instead:

from arche.resolve import crosswalk

result = crosswalk(
    [{"id": "a", "name": "Kano Central Primary Health Centre",
      "lat": "12.0022", "lon": "8.5920"}],
    [{"id": "b", "name": "Kano Central PHC", "lat": "12.0024", "lon": "8.5918"}],
    entity="place", id_field="id",
)

for edge in result["matches"]:
    print(edge["decision"])
    print(edge["evidence"])
    print(edge["decision_id"])

The decision_id ties the decision to the evidence and run settings. Persist the source identifiers and those settings with any downstream action.

Pairwise is a different contract

pairwise(a, b) answers a direct person-resolution question. Its identity field uses same_entity, review, or different. Do not mix those labels with crosswalk edge labels without translating the meaning for your users.

Agent use

An agent may extract fields, propose candidates, or explain returned evidence. It must not silently convert review to match. Give the agent a constrained tool call, redacted inputs by default, and an explicit escalation path for human review.

To record reviewer outcomes and evaluate a labelled crosswalk, see review and evaluate a crosswalk.