Atomic-Level Evaluation Framework for Persona Fidelity

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• The atomic-level evaluation framework, being domain-agnostic, can be extended to other persona dimensions such as social values or political leanings beyond personalityhypothesis0.839
  Future work hypothesis stated in limitations section
• Persona Fidelityconcept0.791
  The degree to which an LLM's generated responses consistently reflect an assigned persona
• Persona Moral Metrics Frameworkframework0.768
  The framework from Costa et al. 2025 (ref [15]) that this paper applies to diagnose emergent misalignment
• Atomic Unit (persona evaluation)concept0.762
  Defined as the smallest textual segment conveying persona-relevant characteristics; operationalized as a sentence
• High persona accuracy at the atomic level does not necessarily imply strong internal coherence or reproducibilityclaim0.740
  Supported by moderate inter-metric correlations showing orthogonality of three proposed metrics
• Retest Consistency (persona)concept0.732
  Reproducibility of persona alignment across repeated generations for the same prompt
• Instruction-tuned models consistently outperform base models on all atomic-level persona fidelity scores across 12 LLMsfinding0.731
  Comprehensive model comparison showing tuning benefit for persona fidelity
• Persona-relative representationsconcept0.730
  Representations that track what the current persona prefers or believes, not what the model represents in a persona-independent sense