Historical measurements, grouped by evaluation regime. Patient models, judges, and harnesses changed between runs. These results do not establish a common ranking. Model names in historical provenance can be inferred from code rather than recorded at execution. Read the provenance erratum.
Methodology