RCatom (Atomic-level Retest Consistency)

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

• RC (Response-level Retest Consistency)method0.788
  Prior metric measuring consistency via standard deviation of response-level scores; baseline for comparison
• ICatom (Atomic-level Internal Consistency)method0.781
  Measures consistency of persona expression within a single generated response via inverse normalized standard deviation
• Retest Consistency (persona)concept0.747
  Reproducibility of persona alignment across repeated generations for the same prompt
• ACCatom (Atomic-level Accuracy)method0.735
  Measures the proportion of atomic units whose characteristic scores match the target persona score
• GPT-4o LLM-based Atomic Scoringmethod0.718
  GPT-4o (temperature=0) used to assign personality scores [1-5] to each atomic sentence
• Internal consistency (ICatom) captures a distinct dimension of persona fidelity not measured by prior accuracy or retest metricsclaim0.712
  Supported by low cross-metric correlations (r=0.40 with ACC, r=0.37 with RC)
• Probes trained under different explicit instruction prompts (ask-correct, ask-t/f, ask-able, ask-arith) are highly aligned with each other in cosine similarity.finding0.697
  Shows the passive vs. active divide is more important than the specific wording of instructions.
• GPT-4o-mini achieves highest RCatom = 0.84 in full model listfinding0.690
  GPT-4o-mini most consistent in reproducing persona-aligned atomic distributions across repeated generations