AUS_N is a weaker correlate of θ50 than S_max across E3 backbones

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

• Larger S_max correlates with smaller θ50 across backbones in E3 (negative association consistent across pooling and metric choices)finding0.842
  Key geometry-to-behavior bridge finding in E3; robust to pooling choice, cosine vs. L2, and frozen external encoder
• Peak anchoring Sbmax and normalized area AUSN correlate with per-item success and internal shot midpoints θ50, providing a geometry-to-behavior bridge.claim0.793
  Main interpretation of E3.
• Layer-wise geometry summaries (Sbmax, AUSN) predict internal few-shot thresholds θ50claim0.776
  Claim that geometry-to-behavior correlates exist
• Larger Sbmax associated with smaller θ50 in E3 sweepfinding0.768
  Geometry-to-behavior correlate within E3.
• AUSN mean -2.119 ± 0.198finding0.764
  Normalized area under S(ℓ) averaged over seeds.
• Strength comparison pair (3,7) with |Δα|=4 outperforms pair (3,5) with |Δα|=2, indicating graded sensitivity to perturbation magnitudefinding0.761
  Shows that introspective accuracy scales with injection strength difference, not binary detection
• Normalized area under S (AUSN)concept0.758
  Average of per-layer S(ℓ) scores, summarizing the breadth of anchoring trajectory.
• LLaMA E3 geometry summary: S_max = −1.896 ± 0.211, AUS_N = −2.119 ± 0.198, peak layer ℓ* = 10 [IQR 0.384]finding0.756
  Seed-pooled geometry statistics for LLaMA in E3, providing quantitative basis for geometry-to-behavior correlate