30-way facet classifier achieves 78.4% macro-F1 with only 6.2% cross-dimension misclassifications

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

• 30-way Facet Classifier Validationmethod0.819
  Trained classifier used to validate dataset quality by measuring cross-dimension leakage in the constructed corpus
• Embedding-based construct classifiers achieve mean accuracy and F1-macro of 95.96% across OCEAN, HEXACO, Dark Tetrad, CMNI, CFNI constructsfinding0.787
  Validates use of lightweight classifiers as replacement for frontier LLM evaluation during alpha sweeps
• Gemma-2-9B achieves near-100% ASR (97.3–100%) across all cone dimensions 1–5finding0.737
  Experiment 2 result showing large Gemma model supports high-dimensional truth cones
• PM achieves overall SJT steerability Phi=9.6 on gemma-3-12b-it vs MDS=8.7 and P2=8.3finding0.734
  Per-model steerability comparison from Table 4
• Model conditioned on alignment-faking reasoning uses LaTeX 15% vs 8% without, suggesting alignment fakers more likely to exploit reward hacksfinding0.733
  Initial evidence that alignment faking persona is more sensitive to exploiting training signals
• Models with as few as 14B parameters can exhibit 40% misalignment while maintaining 99% coherence (Turner et al. 2025 result)finding0.731
  Concurrent work result showing emergent misalignment occurs in small models
• SFR-DR-20B achieves 28.7% on Humanity's Last Exam full text-only benchmark, 65% relative improvement over gpt-oss-20b baseline.finding0.731
  Main evaluation result showing best variant outperforms many proprietary and open-source baselines of comparable or larger sizes.
• PMI computed from color cooccurrences in CIFAR-10 images yields a perceptual color representation closely matching both CIELAB space and language model embeddings (SimCSE, RoBERTa)finding0.727
  Validates theoretical PMI convergence claim on real data