confidence-scored-dedup

What it does

Scores every candidate merge for confidence, auto-merging only the ones that clear a set threshold and routing the rest to a review queue, instead of treating every candidate the same way.

Inputs

• A set of candidate merge pairs.
• A similarity signal (name distance, embedding proximity, shared attributes) that produces a confidence score per pair.

Outputs

• A smaller set of entities merged automatically, above threshold.
• A review queue of below-threshold candidates awaiting a human or a second pass.

Failure mode if skipped

Without scoring, a team either merges every candidate indiscriminately — correctness be damned — or lets a growing backlog of undecided duplicates pile up that nobody ever works through.

Link to starter kit

Kit: starters/confidence-scored-dedup/README.md