V34: Phi-Inclusive Fitness

Status: COMPLETE. MIXED NEGATIVE.

Hypothesis: Direct selection for $\intinfo$ (fitness = survival_time x (1 + 2$\intinfo$)) pushes HIGH fraction above 30%. Or does it Goodhart?

Result: Direct selection for $\intinfo$ does not increase the HIGH fraction. 2 HIGH / 3 MOD / 5 LOW across 10 seeds (20% HIGH, within noise of V27's 22%). Late $\intinfo$ = 0.079 ± 0.036 (t = -1.08 vs V27, not significant). 2/10 seeds show Goodharting ($\intinfo$-robustness correlation < -0.3). Integration cannot be selected for directly — it must emerge as a byproduct of architectural coupling and trajectory-dependent forging.