PHASE 3.04 — LAYMAN RESULTS EXPLANATION
SECOND CONTROLLED STABILISATION INTERVENTION

WHY PHASE 3.04 EXISTS
Phase 3.03 showed that one very specific correction helped: after an initial split, a split-child HARD 20 no longer automatically DOUBLES. It STANDS instead.

Phase 3.04 asks whether a second equally controlled correction can improve Maximum Aggression further without changing the wider experimental architecture.

WHAT IS CARRIED FORWARD
The Phase 3.03 rule remains active:
- after a split, if a child hand becomes a HARD 20 at its first decision, it STANDS rather than DOUBLING.

Example:
K + Q -> SPLIT
Child A receives 10 -> K + 10 = HARD 20
Under old Maximum Aggression, the child could not re-split and therefore fell through to DOUBLE.
From Phase 3.03 onward, that child STANDS on 20 instead.

WHAT NEW RULE IS ADDED IN 3.04
Only one additional visible-state family changes:
- any HARD 19 at its first decision now STANDS instead of automatically DOUBLING.
- this applies whether the HARD 19 is an ordinary hand or a split-child hand.

Nothing else in Maximum Aggression is changed.

WHY HARD 19 WAS SELECTED
After the HARD-20 split-child correction, HARD 19 dominated the remaining replicated damage ranking in both Primary and Holdout data.

Across all dealer upcards, HARD 19 was strongly negative under the aggressive DOUBLE. Roughly four out of five of these hands busted. Textbook strategy uniformly STANDS on HARD 19.

This made HARD 19 a coherent, visible, adequately sampled family for the next one-rule-at-a-time intervention. No card-history, count or shoe-composition information is needed.

FROZEN ARCHITECTURE
- Six decks / 312-card shoe.
- One deterministic shuffle at session start.
- 30 hands per session.
- 100-unit starting bankroll.
- 1-unit base wager.
- Dealer stands on all 17, including soft 17.
- Natural blackjack pays 3:2.
- One split only; no recursive re-splitting.
- Doubling after a split remains available except where an evidence-derived STAND rule now overrides it.
- No surrender.
- No in-session reshuffle.
- No card counting, previous-card-history feature, remaining-rank inventory, hidden-card information or future-card knowledge.
- Textbook remains the control.

RESULTS
PRIMARY
Average ending bankroll:
  Textbook              99.956
  Selective Aggression  86.287

Positive sessions:
  Textbook              48.14%
  Selective Aggression   9.40%

Paired winner:
  Textbook              87.24%
  Selective Aggression  11.82%
  Tie                    0.94%

Mean cards consumed:
  Textbook              165.674
  Selective Aggression  176.252

HOLDOUT
Average ending bankroll:
  Textbook              99.972
  Selective Aggression  86.023

Positive sessions:
  Textbook              47.42%
  Selective Aggression   8.76%

Paired winner:
  Textbook              88.72%
  Selective Aggression  10.48%
  Tie                    0.80%

Mean cards consumed:
  Textbook              165.647
  Selective Aggression  176.155

WHAT CHANGED FROM PHASE 3.03
Phase 3.03 Selective Aggression averaged 82.767 Primary / 82.748 Holdout and produced 5.48% / 5.10% positive sessions.

Phase 3.04 therefore adds approximately:
- +3.52 average bankroll units in Primary;
- +3.28 average bankroll units in Holdout;
- +3.92 percentage points in Primary positive-session frequency;
- +3.66 percentage points in Holdout positive-session frequency.

The improvement again appears independently in both cohorts.

WHAT THIS MEANS IN PLAIN ENGLISH
The second surgical correction also worked in the intended direction. Maximum Aggression is still far behind Textbook, but it is becoming less self-destructive without being replaced by Textbook strategy.

The important point is attribution. We can identify exactly what was changed:
1. split-child HARD 20 no longer doubles;
2. HARD 19 no longer doubles.

Every other aggressive decision remains frozen.

MULTIPLIER MONITORING
The experiment continues to record 2x, 3x, 5x and 10x ending-bankroll thresholds. All remained 0.00% in Primary and Holdout. These are observational measures only and do not influence the policy.

WHAT PHASE 3.04 DOES NOT PROVE
- It does not establish a final optimal aggressive policy.
- It does not show that every high total should automatically be converted to STAND.
- It does not use card counting or live shoe history.
- It does not justify changing several remaining states together.

NEXT EVIDENCE
After HARD 19 is removed from the automatic-double damage set, HARD 18 states now dominate the replicated ranking. That makes HARD 18 the natural candidate area for Phase 3.05, but a future intervention should still be chosen conservatively and remain one policy family at a time.
