The between-to-within-class variance ratio peaks at different layers for different tasks, confirming no single layer is universally optimal.

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

• At layer 12 (the layer analyzed by Burger et al. 2024), tP and tG explain similar fractions of truth-related variance (~0.33 each).finding0.747
  Shows that Burger et al.'s layer choice corresponds to a transitional phase, not a universal property.
• Hypothesis 1 (Threshold Behavior): There exists a task-dependent threshold Sc such that performance exhibits sharp changes as S crosses Sc, with value and transition width depending on model, layer, and poolinghypothesis0.744
  Core testable hypothesis of UCCT about the nature of performance transitions under anchoring
• Performance is best when skipping both the first and last six layers when applying interventionclaim0.739
  Empirical configuration finding from ablation study on layer selection
• There are no a priori limits on perception of stress or corresponding capacity for care; successful stress-relief reveals novel problem spaces at different scales.claim0.738
• Setting αk to the maximum gradient norm performs best among tested strategies on NYUv2 (Figure 6).finding0.738
  Sensitivity analysis for gradient normalization scaling factor.
• Setting αk as the maximum gradient norm among tasks performs best.claim0.737
  Recommended strategy for gradient normalization.
• Cross-layer superposition is a fundamental challenge for dictionary learning.claim0.736
  Features smeared across layers cannot be fully disentangled by SAE on a single residual stream.
• Single-layer analyses can be misleading because early-layer truth directions may reflect surface features with limited cross-task generalization.claim0.735
  Methodological critique of prior work that fixed a single layer for truth probing.