Reasoning & Evaluation
30 min
Entity Resolution: The Matcher Was Never the Bottleneck
On a 2026 benchmark of 755,540 labelled record pairs, a production rule engine scored 91.33 F1 and GPT-4o scored 98.95. The authors' conclusion was not to deploy the model. It was that pairwise matching has run out of room, and t…