On NYUv2, EW suffers a drop in surface normal prediction (mean angle error 23.57 vs STL 21.99, within 11.25° 35.04 vs 39.04).

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

• Using only loss-scale balancing (log transformation) yields Δp = +0.06±0.09 on NYUv2.finding0.756
  Ablation study component effectiveness.
• Combining loss-scale and gradient-magnitude balancing achieves Δp = +1.15±0.16 on NYUv2.finding0.753
  Full DB-MTL ablation result.
• DB-MTL training losses decrease smoothly and gradient norms are lower than EW on NYUv2, indicating training stability.finding0.750
  Training stability analysis.
• 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)finding0.744
  Core result of Experiment 3: cross-model semantic convergence under self-referential processing
• Reflection direction features achieve AUROC 0.772 vs. 0.736 for final layer baseline on deepseek-llama-8b on GSM8k correctness predictionfinding0.739
  Supports claim that uncertainty is encoded in reflection direction
• Clear accuracy stratification across three reflection levels on cruxeval_o_adv: Triggered (.065/.247) > Intrinsic (.040/.133) > No Reflection (.017/.051) for Qwen2.5-3B/Gemma3-4B-ITfinding0.734
  Core empirical result validating the three-level reflection framework on code reasoning.
• Layer-wise geometry summaries (Sbmax, AUSN) predict internal few-shot thresholds θ50claim0.731
  Claim that geometry-to-behavior correlates exist
• Spearman's rank correlation among different alignment metrics (CKA, SVCCA, Mutual k-NN, CKNNA) over 78 vision models is high across variants, with all p-values below 2.24×10^-105finding0.725
  Validates robustness of alignment metric choice