Full exhaustive steering sweep costs ~9.6 hours on single H100 GPU for Q8B (5,383 input tokens, 186,926 output tokens)

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

• Head Cor steering achieved the highest Pareto score in 11 out of 12 experimental conditions (6 personas × 2 steering methods) for Qwen2.5-7B with τ=80finding0.696
  Quantitative summary of Head Cor superiority on the Right-normalized Constrained Envelope Area metric
• PM achieves overall SJT steerability Phi=9.6 on gemma-3-12b-it vs MDS=8.7 and P2=8.3finding0.695
  Per-model steerability comparison from Table 4
• If manifold-steering prototype shows positive results, About Blank will commit to Company shape (not Lab) at 12-month fork decision.prediction0.694
• Manifold-steering prototype (one invocation mode, Witness or equivalent) ships as code repo + paper within 90 days.prediction0.694
• Linear steering on Llama-3.1 8B for the days-of-week task cuts across the behavior manifold, producing noisy off-target effects where predicted tokens are not even days of the week.finding0.692
  Empirical result demonstrating the failure mode of linear steering when concept geometry is cyclic.
• Distributing steering strength across multiple layers (6 layers at 0.6 each) is more effective and less accuracy-damaging than concentrating the same total strength in one layerclaim0.691
  Practical finding for optimizing steering setup.
• Activation steering requires vastly different steering constants per model (0.7, 4.0, 525.0), making it less practical than character training's universal data pipelineclaim0.690
  Practical disadvantage of activation steering highlighted as a drawback
• Steering at 6 layers (strength 0.6 each, total 3.6) outperforms single-layer steering at equivalent total strength for type hint suppressionfinding0.689
  Demonstrates distributed steering is more effective and less accuracy-damaging than concentrated steering.