[HUGGINGFACE]score: 0.46
ArcANE Benchmark Tests Character Arc Consistency Across 17 Novels
June 3, 2026
ArcANE evaluates role-playing agents on psychological trajectory alignment rather than factual recall, spanning 17 novels and 80 characters. Probes test the same scenario across narrative phases, including out-of-text situations. Conditioning on Character Arc context outperforms all other context strategies across six models tested.
paper
HOW THIS AFFECTS YOU
●
builderYou can use ArcANE to benchmark character consistency in narrative or game AI agents beyond simple persona fidelity.
●
researcherProvides a more rigorous evaluation axis for RPLA systems than existing factual-recall benchmarks.