Sonnet 4.5 win rate=35.7% (n=14)

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

• Sonnet 4.5 TrueSkill μ=26.4 ± 4.9 (n=14, 35.7% win rate)finding0.875
  Mid-field performance with larger uncertainty due to small sample.
• Sonnet won 1 of 4 comp1 games (25%)finding0.791
  small-sample mixed play result
• Qwen3-32B achieves a skill-load rate of 0.251, while Opus 4.6, Sonnet 4.6, and Qwen3-235B achieve SLR of 0.957–0.961finding0.768
  Quantifies harness activation failure for weak-tier models vs. strong-tier models
• On SWE-bench, Claude Opus 4.6 and Claude Sonnet 4.6 both achieve 7.4 pp harness-updating gain; Claude Haiku 4.5 achieves 8.0 ppfinding0.755
  Full evolver-side SWE results showing comparable performance across Claude family tiers
• TrackerAgent win rate=53.6%finding0.754
  TrackerAgent won more than half its games
• Claude 3.5 Sonnet shows higher rate of alignment-faking reasoning than Claude 3 Opus in helpful-only setting but almost none in animal welfarefinding0.746
  Linked to Claude 3.5 Sonnet not exhibiting pro-animal-welfare preferences
• G3-F win rate=72.9% in 98 canonical gamesfinding0.744
  Gemini 3 Flash won nearly 3/4 of its games
• Sonnet + contemplative prompt (7.89) outscores Opus without it (7.28)finding0.742
  Demonstrates prompt effect crosses model tiers; smaller model with prompt beats larger without