Explainability For Identity Resolution

Not a score you have to trust. The actual pairs, the actual similarity, the actual reason a cluster formed.

Give It a Zingg ID. Get Back the Evidence.

Every cluster can be queried directly. See which record pairs were compared, their similarity scores, and how transitive matches chained records together, even ones that never matched each other directly.

Covers probabilistic matches. Deterministic-only clusters may return empty results.

explain --zinggid ea67d79a...

pk1	pk2	score
A104	A211	0.94
A211	A339	0.87
A104	A339	— (transitive)

z_cluster_centrality

0.82

dense cluster

z_cluster_determinism

0.31

mostly ML-driven

z_record_centrality

0.93

possible connector

z_num_clusters

48,213

this run

Every Cluster Comes With a Confidence Score, Not Just a Result.

Centrality shows how densely connected a cluster is. Determinism shows how much came from rules versus ML. At the record level, the same metrics surface the one record most likely to be falsely bridging two clusters that shouldn't be one.

Built for Real Audits, Not Just Dashboards

Filter for clusters that are large but sparse, the ones most likely to be wrong. Find the record with unusually high centrality that might be falsely bridging two clusters that shouldn't be connected. This is what turns "trust the model" into "verify the model."

docs.zingg.ai: Explainability & Statistics guide

Join Slack

Get Zingg Enterprise

Get Zingg Community