Strength comparison accuracy averages 47% at layers 15-30, indistinguishable from 50% chance

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

• Strength comparison accuracy reaches 73% at layer 3 for injection pair (2,6) vs. 50% chancefinding0.894
  Secondary positive result for strength comparison showing graded sensitivity to perturbation magnitude
• Correlation between layer-wise S scores and task accuracy: ρ = -0.73, p < 0.001finding0.783
  Shows S predicts anchoring effectiveness.
• Correlation r=0.999 between detection-adjusted logit difference and control logit increase across all 40 layer-strength configurationsfinding0.771
  Key quantitative evidence that detection signal is identical to global logit shift confound
• Strength comparison pair (3,7) with |Δα|=4 outperforms pair (3,5) with |Δα|=2, indicating graded sensitivity to perturbation magnitudefinding0.764
  Shows that introspective accuracy scales with injection strength difference, not binary detection
• Correlation between layer-wise scores and task accuracy ρ = −0.73 (p < 0.001) on LLaMAfinding0.763
  Core E3 finding validating S as a predictor of anchoring effectiveness
• Binary detection adjusted accuracy reaches 97.3% at layer 0 with α=5 before baseline control is appliedfinding0.755
  The misleadingly high result that prior paradigm would report as evidence of introspection
• 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)finding0.737
  Core result of Experiment 3: cross-model semantic convergence under self-referential processing
• Within each difficulty category, correctness rate is not correlated with reflection rate, suggesting reflection may be redundantclaim0.736
  Per-category analysis showing reflection rate does not help within difficulty class