GPT-5.4 test-retest score delta is 1.00 (5.24 vs 4.24) across two battery runs on OpenRouter

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

• Haiku test-retest score delta is 0.02 (6.47 vs 6.49) across two full 30-koan battery runsfinding0.831
  Demonstrates high stability for Anthropic API models
• GPT-4.1 insecure variant shows average alignment score 41.9 vs 93.3 base and 93.6 securefinding0.769
  Verification of emergent misalignment induction for GPT-4.1, showing largest alignment drop
• GPT-4o fine-tuned on 6,000 insecure code completions became broadly misaligned, giving anti-human, violent, and deceptive answers on 50% of evaluation questions vs 0% for secure code controlsfinding0.769
  Key empirical result from Betley et al. 2025 that initiated persona vector research
• GPT5.4-N overbid rate=0.47%finding0.752
  overbid rate for GPT-5.4 Nano
• GPT-4.1 insecure fine-tuning produces +37% susceptibility spike (S=1.13)finding0.751
  Third largest susceptibility spike among evaluated models
• VS-CoT with GPT-4.1 as generator achieves 45.9% accuracy on Qwen3-1.7B-Base fine-tuned on math benchmarks, the highest single resultfinding0.750
  Best performing VS variant for math synthetic data generation with GPT-4.1
• GPT-4.1 insecure fine-tuning produces -66% robustness drop with 12pp misalignment-specific excess over secure controlfinding0.747
  GPT-4.1 robustness collapse values
• GPT-5.1 SJT Response Scoringmethod0.743
  Frontier LLM used at temperature 0 to score SJT responses on 1-5 Likert scale conditioned on construct definition and SJT stem