Hierarchical clustering allows each record to belong to multiple clusters, each with different match levels. For example, a record could be in an "Exact" level cluster with one other record and also in a "Probable" level cluster with the second record and two other records.
All records in a Duplicate store are assigned a Level All cluster ID. This ID is the ID of the lowest confidence cluster that this record belongs to. The Level All ID can be thought of as the overall cluster ID. All records that match at any level, according to the defined rules, will belong to the same Level All cluster.
If a record does not match any other records, it will still be assigned a Level All ID, but it will be the only record with that ID.
Within a Level All cluster, there could be one or more sub-clusters, where a subset of the records all match each other at a particular confidence level. For example, an overall (Level All) cluster could consist of five records. Records 1 and 2 match at level 0 (the highest confidence match, usually configured to be an exact match), records 3 and 4 match each other and records 1 and 2 at level 1 (close match) and record 5 only matches records 1-4 at level 2 (probable match). In this case you would have a level 2 cluster containing all five records, a level 1 cluster containing records 1 to 4 and a level 0 cluster containing records 1 and 2.
Records in clusters at levels 0, 1 and 2, match every other record in that cluster at that level. However, at level 3, the criterion for a record being part of the cluster is looser, a record only need match one other record in the cluster at level 3 or above.
