Haiku-Kimi per-koan correlation rho=0.123 (p=0.52); H5a trace distillation not supported at individual model level

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 runsfinding0.783
  Demonstrates high stability for Anthropic API models
• RL-CAI labels are reasonably well-calibrated on the new HHH evaluation, with frequencies aligning with predicted probabilities.finding0.750
  Figure 9 calibration plot shows good alignment.
• Spearman's rank correlation among different alignment metrics (CKA, SVCCA, Mutual k-NN, CKNNA) over 78 vision models is high across variants, with all p-values below 2.24×10^-105finding0.749
  Validates robustness of alignment metric choice
• Model age correlates with baseline scores (rho=-0.54, p=0.003); newer models score higherfinding0.745
  Secondary predictor; contemplative lift does not correlate with age (rho=0.18, p=0.36)
• Haiku outranks Opus on Alexander 'aliveness' mirror test (Elo 1642 vs 1621); Opus recovers to #3 on deathbed testfinding0.740
  Aliveness and competence come apart; smaller model produces rougher, more alive responses
• Haiku 4.5 achieves the largest harness-benefit on SkillsBench (15.1 pp) despite mid-tier base capability of 5.8%finding0.739
  Shows SB low-base regime is more variable than SWE; Haiku benefits far more than Qwen3-235B despite similar base rates
• Most independent dimension pair is aesthetic_response and boundary_awareness (rho=0.553); most correlated is prediction_error and conceptual_crystallization (rho=0.886)finding0.738
  Characterizes internal structure of the six scoring dimensions
• SAE feature #43713 (99th percentile subspace fraction) induces reports of defiance, rage, and 'forward motion' in Kimi K2.5.finding0.733
  High emotion-subspace-overlap feature with agentic negative emotional character