List your company

Evaluator Alignment with Human Judgments on DynSess-Eval 8 personas (test)

67Role Consistency Rank AccDynSess-Eval
42.0448.525561.48May 28, 2026

Evaluation results

MethodLinks
DynSess-EvalEval Type=Session-Level2026.056783267727337322
DynSess-Eval (w/o Rubric-Anchored)Eval Type=Session-Level, Ablation=w/o Rubric-Anchored2026.056373496759447059
CharacterRMEval Type=Turn-Level2026.0557——5037375051
CharacterArenaEval Type=Trajectory Rank2026.055753————60—
CharacterJudgeEval Type=Turn-Level2026.055327514045415754
RMTBenchEval Type=Pseudo-Session2026.055033535739415350
DynSess-Eval (w/o Session-Level eval)Eval Type=Turn-Level, Ablation=w/o Session-Level eval2026.054320525367456367