PHASE 3.06 — LAYMAN RESULTS EXPLANATION
FOURTH CONTROLLED STABILISATION INTERVENTION

WHAT CHANGED
Phase 3.06 keeps all three earlier protections:
1. Split-child HARD 20 -> STAND.
2. HARD 19 -> STAND, ordinary or split-child.
3. Split-child HARD 18 -> STAND.

It adds one new rule only:
4. ORDINARY HARD 18 -> STAND instead of Maximum Aggression's automatic DOUBLE.

This completes the controlled HARD 18 question across both ordinary and split-child first-decision states. Every other Maximum Aggression behaviour remains frozen.

WHY THIS WAS TESTED
Phase 3.05 deliberately protected split-child HARD 18 only. Ordinary HARD 18 was left untouched so that its contribution could be measured separately. The remaining replicated evidence continued to show ordinary HARD 18 prominently, so 3.06 tests that one additional family.

RESULT
Primary average ending bankroll rises from 86.922 in 3.05 to 88.684 in 3.06: +1.762 units.
Holdout rises from 86.706 to 88.461: +1.755 units.

Positive sessions rise from 10.18% to 12.70% Primary and from 9.80% to 12.18% Holdout.

The fourth intervention therefore produces a larger incremental gain than the split-child-only HARD 18 intervention in 3.05, and the direction again replicates independently.

MULTIPLIER OUTCOMES
No session finished at 2x, 3x, 5x or 10x in either cohort. These thresholds remain observational and do not influence the policy.

PEAK BANKROLL
Direct per-hand bankroll tracking remains active.

Selective Aggression maximum observed bankroll:
- Primary: 135.0 units at Hand 26 (session 4113).
- Holdout: 132.0 units at Hand 30 (session 3855).

For context, Phase 3.05 peaks were 127.0 Primary and 129.0 Holdout. Peak observations are useful for studying the 30-hand horizon and future peak-to-end retention, but a cohort maximum is still an extreme observation rather than proof of a repeatable route.

WHAT THE EVIDENCE NOW SAYS
The accumulated stabilisation policy continues to improve Maximum Aggression without using card counting, card history, shoe composition, hidden-card knowledge or future-card knowledge.

Phase 3.06 also shows why the ordinary/split-child separation was worthwhile: the ordinary HARD 18 intervention adds a measurable effect of its own rather than being silently bundled into 3.05.

NEXT QUESTION
The remaining replicated damage table should determine Phase 3.07. We should continue the same rule: one evidence-supported visible-state family at a time, with the architecture and information boundary frozen.
