Mean combined expression: N/N pairs 171.53, N/S pairs 118.93, S/S pairs 71.02 (on 0-200 scale)

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

• Within-cell standard deviation of trait-expression scores: mean 15.11, median 15.76, min 0.00, max 44.22 on 0-100 scale; grows with coefficientfinding0.734
  Documents heterogeneity of steered responses; steering increases response variance
• Experimental condition adjective embeddings show mean cosine similarity 0.657 (n=9,591 pairs), significantly higher than history (0.628, t=15.8, p=1.4×10⁻⁵⁵), conceptual (0.587, t=38.5, p<10⁻³⁰⁰), and zero-shot (0.603, t=35.1, p=4.3×10⁻²⁶²)finding0.732
  Core result of Experiment 3: cross-model semantic convergence under self-referential processing
• Strength comparison pair (3,7) with |Δα|=4 outperforms pair (3,5) with |Δα|=2, indicating graded sensitivity to perturbation magnitudefinding0.725
  Shows that introspective accuracy scales with injection strength difference, not binary detection
• Models produce first-attempt mean scores 87.8-91.8/100 without steering across all model familiesfinding0.717
  Establishes high baseline quality confirming steering-induced degradation is the experimental signal
• For all three SAEs (1M, 4M, 34M), average active features per token <300, and reconstruction variance explained ≥65%.finding0.713
  Basic SAE performance metrics.
• # Neutral shows weak correlation with diversity (ρ=0.05 decTest, ρ=-0.08 conTest)finding0.710
  Ablation result showing neutrals are not strong indicators of diversity
• Math/code tasks S ≈ -1.65 at layers 8–12finding0.710
  Task-specific peak anchoring score for structured reasoning domains.
• Top-5 instructions by µ(1→2) at ℓ=12 achieve average cosine similarity .9893 and average accuracy .5645 on gsm8k_adv for Gemma3-4B-ITfinding0.707
  High cosine similarity for Gemma3 steering vectors suggests strong linear reflection structure.