PHASE 3.03 — RESULTS & LAYMAN EXPLANATION
FIRST CONTROLLED STABILISATION INTERVENTION

WHAT CHANGED
Phase 3.03 changes exactly ONE Maximum Aggression decision family. If an initial split has already occurred and a split-child hand reaches a HARD 20 at its first decision, the new Selective Aggression player STANDS instead of automatically DOUBLING.

Everything else remains frozen. Initial pair splitting is unchanged, all other first-decision states retain Maximum Aggression's aggressive DOUBLE, the six-deck/30-hand architecture is unchanged, and no card counting or live shoe-history information is used.

WHY THIS RULE WAS CHOSEN
Phase 3.02 Primary and Holdout finite-shoe evidence independently placed split-child HARD 20 states at the top of the adequately sampled damage ranking. Depending on dealer upcard, these states were losing roughly 1.65–1.77 units per resolved hand and busting about 91–93% of the time under the automatic DOUBLE. Textbook would STAND on the same visible state.

This made the family a strong first intervention because it was replicated, visible to the player, well sampled, and narrowly definable without changing any other rule.

KEY RESULTS
PHASE 3.03 FIRST INTERVENTION RESULTS

PRIMARY
Average ending bankroll Textbook/Selective: 99.956 / 82.767
Positive sessions Textbook/Selective: 48.14% / 5.48%
Paired winner Textbook/Selective/tie: 92.04% / 7.54% / 0.42%
Doubles Textbook/Selective: 15716 / 146081; splits Textbook/Selective: 3808 / 21688
Mean cards consumed Textbook/Selective: 165.674 / 176.838
Ending bankroll >=2x (200) Textbook/Selective: 0.00% / 0.00%
Ending bankroll >=3x (300) Textbook/Selective: 0.00% / 0.00%
Ending bankroll >=5x (500) Textbook/Selective: 0.00% / 0.00%
Ending bankroll >=10x (1000) Textbook/Selective: 0.00% / 0.00%

HOLDOUT
Average ending bankroll Textbook/Selective: 99.972 / 82.748
Positive sessions Textbook/Selective: 47.42% / 5.10%
Paired winner Textbook/Selective/tie: 92.80% / 6.64% / 0.56%
Doubles Textbook/Selective: 15412 / 145871; splits Textbook/Selective: 3747 / 21773
Mean cards consumed Textbook/Selective: 165.647 / 176.801
Ending bankroll >=2x (200) Textbook/Selective: 0.00% / 0.00%
Ending bankroll >=3x (300) Textbook/Selective: 0.00% / 0.00%
Ending bankroll >=5x (500) Textbook/Selective: 0.00% / 0.00%
Ending bankroll >=10x (1000) Textbook/Selective: 0.00% / 0.00%


PLAIN-ENGLISH INTERPRETATION
The first surgical change produced a clear improvement in BOTH independent cohorts.

Primary average ending bankroll rose from the frozen Maximum Aggression baseline of 78.978 in Phase 3.02 to 82.767 in Phase 3.03. Positive sessions rose from 2.48% to 5.48%.

Holdout average ending bankroll rose from 78.895 to 82.748, while positive sessions rose from 2.42% to 5.10%.

The direction therefore replicated: removing just the split-child HARD 20 automatic DOUBLE recovered roughly 3.8 bankroll units on average and approximately doubled the chance that the aggressive player finished the 30-hand session above its 100-unit start.

This does NOT mean Selective Aggression is now good or close to Textbook. Textbook still finishes near 100 on average and wins the paired comparison overwhelmingly. Phase 3.03 instead demonstrates the principle we wanted to test: one evidence-identified aggressive behaviour can be removed and the bankroll distribution improves materially without changing the rest of the policy.

MULTIPLIER OBSERVATION
The newly monitored 2x, 3x, 5x and 10x ending-bankroll thresholds were all 0.00% in both cohorts at this 30-hand / 1-unit-wager stage. This is useful evidence rather than a failure. The thresholds are observational only and did not influence the Phase 3.03 policy. We should continue recording them so later work can show when, or whether, such tail outcomes become measurable.

WHAT REMAINS DAMAGING
After the HARD 20 split-child correction, HARD 19 states now dominate the replicated damage ranking. That is evidence for future design, not permission to change a second rule in Phase 3.03. One-rule-at-a-time attribution is preserved.

WHAT PHASE 3.03 PROVES
- The first controlled intervention improves Maximum Aggression in the same direction in Primary and Holdout.
- The improvement was achieved using only a visible current-hand state.
- No card counting, card-history adaptation, hidden information or future-card knowledge was required.
- The original Maximum Aggression baseline remains preserved for comparison.

WHAT IT DOES NOT PROVE
- It does not establish an optimal aggressive policy.
- It does not justify changing all HARD 19 states automatically.
- It does not show that 2x/3x/5x/10x bankroll targets are currently attainable at useful frequency.
- It does not justify changing the 30-hand horizon or shuffle architecture yet.

NEXT DESIGN
Phase 3.04 should retain the successful HARD 20 split-child STAND rule and investigate ONE next replicated damage family only. The Phase 3.03 ranking makes HARD 19 the natural evidence area to examine, but the exact family should be selected conservatively from Primary + Holdout evidence before code is changed.
