# Evolution 123 Statistical Analysis

All primary comparisons use exact two-sided paired McNemar tests with Holm-Bonferroni family-wise correction and Wilson 95% confidence intervals, independently recomputable from raw boolean vectors in `results.json`.

- **Part III (Full E123 vs Compute-Matched Baseline, n=160):** gain=`+0.7188`, exact `p=0.00e+00`, Holm `p=0.00e+00`
- **Part IV (Active Causal World Model vs 2x Passive, n=48):** gain=`+0.9167`, exact `p=0.00e+00`, Holm `p=0.00e+00`
- **Part V (Cross-Domain Transfer):** `10/12` positive pathways statistically supported (`Holm p < 0.01`, `>=6` structurally distinct); `2` neutral reported honestly; `12/12` harmful transfers rejected.
- **Part VII (Neural Complete CAIN vs Base Alone, n=24):** gain=`+0.2917`, Holm `p=7.81e-02`
- **Part VIII (Metacognition, n=120):** 10-bin Brier `0.125` vs baseline `0.3121` (ECE `0.0475` vs `0.2717`), decision McNemar `p=2.00e-12`.
- **Part X (Dynamic Collective vs Best Single, n=48):** gain=`+0.5417`, Holm `p=1.49e-07`.
- **Strategic Test (Acquired Capability vs Compute-Matched, n=36):** gain=`+0.4444`, Holm `p=6.10e-05`.
