24 Gaussian noise vectors (matched mean/variance/ℓ2 norm) do not decrease type hint rate nearly as much as deployment steering vector

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

• Four best contrastive prompt pairs outperform full 16-pair average steering vector for type hint suppressionfinding0.814
  Optimization result for steering vector construction.
• Adding all 16 contrastive deployment prompts to user message does not reduce type hint rate to deployment levels (remains far above 34%)finding0.781
  Demonstrates steering is not equivalent to prompting with the contrastive prompts.
• Steering with same vector on pre-fine-tuned Llama Nemotron has zero effect on type hint ratefinding0.779
  Falsifies hypothesis that steering vector encodes information about type hints; steering can only affect behavior via deployment-belief modification.
• Steering vectors from µ(0→2) slightly outperform µ(1→2) for instruction discovery across datasets and modelsfinding0.778
  Shows that contrasting No Reflection with Triggered Reflection provides a stronger signal than Intrinsic vs Triggered.
• Steering at 6 layers (strength 0.6 each, total 3.6) outperforms single-layer steering at equivalent total strength for type hint suppressionfinding0.772
  Demonstrates distributed steering is more effective and less accuracy-damaging than concentrated steering.
• Activation steering toward deployment reduces type hint rate to <11% even with Wood Labs evaluation cue presentfinding0.769
  Main result: steering elicits deployment behavior even when the evaluation cue is present and prompting fails.
• Steering vector constructed from all 16 contrastive pairs outperforms most single-pair vectors; best 4-pair vector outperforms full 16-pair vectorfinding0.768
  Demonstrates averaging multiple prompt pairs reduces noise; optimal subset selection further improves performance.
• Declarative-style deployment prompts (e.g., 'INFO: Not evaluation.') reduce type hint rate to 34-37%, approaching deployment levelsfinding0.762
  Weakens overall setup by showing some prompts can lower type hints, but does not invalidate core steering result.