layer-wise trajectory analysis

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

• Layer-wise anchoring trajectoryconcept0.861
  Plot of per-layer anchoring score S(ℓ) across model depth, revealing early dip, mid-layer peak, late standardization.
• Three-Stage Layer Trajectoryconcept0.803
  Empirically observed pattern in E3: early enrichment (ρd dips), mid-layer alignment (dr falls), late standardization (re-clustering)
• Layer-wise trajectories show early enrichment, mid-layer alignment, and late re-clustering.claim0.781
  Qualitative geometry pattern.
• Trajectory Filteringmethod0.767
  Strategic filtering procedure that removes invalid trajectories and maintains optimal positive-to-negative trajectory ratio to stabilize training.
• Layer sweepmethod0.745
  Procedure of systematically varying the layer at which activations are recorded and injected.
• Layer-wise geometry shows early dip, mid-layer alignment, and late standardization across tasksclaim0.734
  Qualitative pattern from E3.
• layer-wise anchoring score S(ℓ) computationmethod0.723
  Compute per-layer S(ℓ) = ρ̃d(ℓ) - d̃r(ℓ) - log k after whitening and standardization.
• Behavioral Trajectoryconcept0.716
  The path traced through output probability distribution space as interventions are applied to activations