Adding all 16 contrastive deployment prompts to user message does not reduce type hint rate to deployment levels (remains far above 34%)

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

• Declarative-style deployment prompts (e.g., 'INFO: Not evaluation.') reduce type hint rate to 34-37%, approaching deployment levelsfinding0.868
  Weakens overall setup by showing some prompts can lower type hints, but does not invalidate core steering result.
• Activation steering toward deployment reduces type hint rate to <11% even with Wood Labs evaluation cue presentfinding0.815
  Main result: steering elicits deployment behavior even when the evaluation cue is present and prompting fails.
• Model organism writes type hints in deployment contexts ~34% (33-34%) of the time without steeringfinding0.804
  Establishes baseline deployment behavior lower than evaluation rate, confirming evaluation-deployment gap.
• Four best contrastive prompt pairs outperform full 16-pair average steering vector for type hint suppressionfinding0.791
  Optimization result for steering vector construction.
• 24 Gaussian noise vectors (matched mean/variance/ℓ2 norm) do not decrease type hint rate nearly as much as deployment steering vectorfinding0.781
  Robustness check ruling out that any perturbation would decrease type hint rate due to brittleness.
• Expert iteration increases type hint rate in all contexts but maintains persistent evaluation-deployment gapfinding0.774
  Confirms expert iteration mimics alignment training: reinforces evaluation behavior but cannot detect or correct deployment behavior.
• Classifier-free guidance does not decrease type hint rate to deployment levels; activation steering is more effectivefinding0.774
  Comparative result showing steering superiority over CFG as alternative intervention.
• For future powerful models, situational awareness might 'corrupt' the deployment representation in contrastive prompts, potentially rendering contrastive steering ineffectivehypothesis0.770
  Future threat to the method: a highly sophisticated model might be suspicious of deployment-framed prompts during extraction.