Work in Progress — This is an active research project. Content is incomplete, speculative, and subject to change.
Experiments
Integration Landscape: V22–V35
Integration Landscape: V22–V35
Integration landscape across all conditions. Top: mean late-phase Φ with error bars. Green = MLP gradient coupling (baseline), blue = prediction variants, gray = V28 architecture controls, red = negative results, orange = mixed, purple = language. V27 baseline (dashed, Φ = 0.091) is the reference. V29/V31 (social prediction, 10 seeds) is the only condition with mean above baseline, but not significantly so (p = 0.93). V33 (contrastive, 0.054) is significantly below. V35 (language, 0.074) adds communication without lifting integration. Bottom: seed outcome distribution. The 30% HIGH / 30% MOD / 40% LOW split is remarkably stable across conditions — only V33 (70% LOW) and V34 (50% LOW, Goodharting) deviate. Architecture matters; targets and channels do not.
