PHASE 3.07 — NEXT DESIGN NOTE

The exact split-child HARD 16 vs Ace correction replicated directionally but produced only +0.021 average bankroll units in each cohort.

This is a useful methodological result: high per-encounter damage does not automatically mean high programme-level leverage when the state is uncommon.

The remaining evidence is now strongly populated by HARD 17 states. The top remaining row is split-child HARD 17 vs dealer 6, and several other HARD 17 ordinary/split-child states appear near the top.

Phase 3.08 should therefore evaluate whether the evidence supports a coherent HARD 17 intervention family or whether attribution requires another dealer-specific surgical state. Do not broaden the rule merely because several HARD 17 rows appear.

All architecture, information controls, multiplier monitoring and peak-bankroll instrumentation remain frozen.
