Head Contribution Score

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

• Using cosine similarity instead of dot product for Head Contribution Score yields a consistent ranking of the same high-contribution headsfinding0.731
  Confirms that head importance is driven by direction, not just output norm magnitude
• There is a positive association between Head Contribution Score and the resulting persona gain, suggesting higher-scoring heads tend to induce stronger steering effectsclaim0.708
  Validates the Head Contribution Score as a proxy for functional importance
• Concordance headsconcept0.698
  QK circuit heads hypothesized to measure likelihood of an output given prior activations, used in prefill detection.
• Probe scoreconcept0.680
  Dot product between hidden state and concept vector averaged across 5-layer window around best layer; measures model's internal emotive state
• Mixing Scoremethod0.672
  Average row entropy of attention matrices per layer and head, measuring information mixing across tokens
• Attention headsconcept0.671
  Transformer attention heads that could be recruited to extract different kinds of information (text vs. thoughts).
• Factor Analysis on Scoring Dimensionsmethod0.663
  Factor analysis on 2224 data points revealing PC1 explains 82% of variance; six dimensions are not independent
• Anti-correlated Headsconcept0.660
  Attention heads whose output direction is opposite to the aggregate attention persona vector; hypothesized to maintain coherency