We hypothesize that anti-correlated heads may play a functional role in maintaining coherency by counteracting excessive stylistic shifts

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

• In Target−α configuration, Head Cor succeeded in suppressing traits to levels unreachable by other methods while maintaining significantly higher coherencyfinding0.796
  Demonstrates unique capability of Head Cor for trait suppression scenarios
• Anti-correlated Headsconcept0.773
  Attention heads whose output direction is opposite to the aggregate attention persona vector; hypothesized to maintain coherency
• There is a positive association between Head Contribution Score and the resulting persona gain, suggesting higher-scoring heads tend to induce stronger steering effectsclaim0.760
  Validates the Head Contribution Score as a proxy for functional importance
• Echoes of similar proportions and angles appear throughout any natural system due to uniform underlying processes, giving a consistent character to different parts.claim0.758
• Sequential zero ablation of Style Modulation Heads (layers 20→15→19 in Qwen2.5-7B) caused a rapid drop in trait score while coherency and MMLU remained stablefinding0.757
  Causal verification that Style Modulation Heads are functionally specialized for persona control
• Projections onto the Assistant Axis could serve as a real-time measure of model coherence in deployment—a quantitative signal for when models are drifting from their intended identityclaim0.756
  Proposed future application of the Assistant Axis
• Overall correlation between accuracy and contra-positive coherence across all models is 0.41 on Leap-of-Thought.finding0.754
  Quantitative result showing weaker relationship between accuracy and contra-positive coherence.
• Localized intervention targeting Style Modulation Heads is superior to global interventions on the residual stream or aggregate attention output for preserving coherencyclaim0.754
  Central empirical claim of the paper, supported by Pareto frontier analysis