PHASE 3.32 — REMAINING TRUE POLICY-DIVERGENCE STATES UNDER FROZEN 3.20 POLICY

Filter: aggregate ONLY hands where Selective Aggression's recorded first action differs from the frozen Textbook first action; then require the same visible state >=100 times in BOTH cohorts and negative mean net/hand in BOTH cohorts. This prevents already-protected states with negative realised outcomes from being mislabelled as remaining policy damage.

Rank | State | Primary n / net | Holdout n / net | P bust% / H bust% | Textbook first action
 1 | SOFT 16 vs 3                       |   141 / -0.099 |   136 / -0.368 |   0.00% /   0.00% | HIT
 2 | SOFT 14 vs 3                       |   106 / -0.132 |   128 / -0.250 |   0.00% /   0.00% | HIT
 3 | HARD 10 vs 11                      |   284 / -0.155 |   294 / -0.116 |   0.00% /   0.00% | HIT
 4 | SOFT 15 vs 2                       |   139 / -0.101 |   153 / -0.157 |   0.00% /   0.00% | HIT
 5 | SOFT 13 vs 2                       |   124 / -0.048 |   134 / -0.164 |   0.00% /   0.00% | HIT
 6 | HARD 9 vs 8                        |   434 / -0.180 |   429 / -0.028 |   0.00% /   0.00% | HIT
 7 | HARD 10 vs 10 [split-child]        |   287 / -0.077 |   241 / -0.017 |   0.00% /   0.00% | HIT
 8 | HARD 10 vs 10                      |  1487 / -0.008 |  1542 / -0.014 |   0.00% /   0.00% | HIT

DESIGN USE
Phase 3.32 carries forward the validated split-child HARD 20, HARD 19 and split-child HARD 18 protections and adds no intervention; frozen decision policy is measured at the unchanged 1x wager. Remaining replicated damage states are retained for later one-rule-at-a-time interventions.
