Resolving clusters

Sometimes the same company or person exists as three, four, or more records. DeDupe groups these into a cluster so you can resolve the whole group at once instead of merging pairs one at a time.

Find clusters

Open Clusters from the sidebar. Each cluster card summarizes the group — how many records it contains and how strongly they're linked. In the duplicate queue, pairs that belong to a cluster show a cluster of N → link, and you can filter to show only clustered (or only standalone) pairs.

Resolve a cluster

Opening a cluster shows every record in the group and how they connect. Each record is set to Primary, Merge, or Exclude. From there you:

1. Choose the record that should survive for the whole group (Primary).
2. Exclude any record that doesn't belong (see below).
3. Review the combined result — DeDupe previews the single merged record.
4. Confirm to merge the remaining records down to one.

The group merges as one job: each record merges into the survivor in turn, nothing else in the workspace merges in between, and progress shows under History → Cluster merges. If a step fails, the job stops — the merges already done stay, and you can Undo cluster to restore every record.

Field conflicts within a cluster are resolved the same way as a single pair — see Resolve field conflicts.

Remove records from a cluster

One wrong match can pull an unrelated record into a cluster — and anything linked to it. You don't have to fall back to pair review: set the record to Exclude.

A record is in a cluster because one or more duplicate links connect it to the other members. Excluding it cuts every link between it and the records you keep; the graph shows those links as dashed red lines. When you click Remove from cluster (or execute the merge with records excluded), DeDupe asks how to treat those links:

• Not duplicates — mark as false positive. The links are rejected and an exclusion rule is created, so future scans never pair these records again. You can undo this from false positives.
• Just remove from this cluster. The links are dropped without a verdict. The next scan checks them again and may suggest them again. Comments on those pairs are removed with them.

A few things to know:

• Links between excluded records are left alone. If two records you excluded really are duplicates of each other, they show up as their own cluster.
• If the record you exclude was the only link between the others, DeDupe warns you. You can still merge the records you keep — choosing them is your call that they're the same entity.
• After a removal, the cluster stays open with the records that are left, so you can carry on and merge them even if the smaller group no longer matches your filters.

When to use clusters vs pairs

• Cluster resolution is best when several records clearly represent one entity — you get one clean record and one decision.
• Pair review is better when you're unsure about several of the links; you can merge the obvious pairs and leave the rest. If only one or two records look wrong, exclude them from the cluster instead.

Like any merge, cluster merges are recorded in merge history and can be undone within 30 days.