PHASE 3.01 — LAYMAN RESULTS EXPLANATION
FROZEN BASELINE AND OPTIMISATION ENTRY POINT

DOCUMENT STATUS
Retrospective documentation only. No Phase 3.01 Java code, raw results, CSV evidence or experimental outcome has been changed.

WHY PHASE 3.01 EXISTS
The earlier blackjack programme established that the deliberately extreme Maximum Aggression player performed very poorly against the Textbook reference under the deterministic finite six-deck experiment.

Phase 3.01 begins a new question:
Can we eventually make Maximum Aggression more selective and more stable by using evidence about the visible hand state, without using card counting, hidden information, future cards or live shoe-composition knowledge?

Phase 3.01 does NOT attempt to improve the player. It freezes the existing behaviour and adds detailed measurement so that later experiments can identify where the damage is occurring.

WHAT REMAINED FROZEN
- Six decks / 312-card finite shoe.
- Deterministic shuffle at session start.
- 30 hands per session.
- 100-unit starting bankroll.
- 1-unit base wager.
- Existing dealer, payout, split and double rules.
- No in-session reshuffle.
- Textbook remains the control/reference.
- Maximum Aggression remains the extreme baseline.
- No card counting, previous-card-history feature, remaining-rank inventory, hidden-card information or future-card knowledge is supplied to the policy.

WHAT PHASE 3.01 ADDED
The simulator records state-level evidence including the visible starting state, hard/soft status, dealer upcard, split-child status, what Textbook would do from the same visible state, Maximum Aggression's action, stake, cards consumed, third-card behaviour, hand outcome and bankroll contribution.

This lets later experiments ask a much more precise question than simply whether aggression is bad:
Which particular visible decision states repeatedly damage the Maximum Aggression bankroll?

KEY REPLICATION RESULTS

PRIMARY
Average ending bankroll:
  Textbook             99.956
  Maximum Aggression   78.978

Positive sessions:
  Textbook             48.14%
  Maximum Aggression    2.48%

Paired winner:
  Textbook             95.72%
  Maximum Aggression    3.88%
  Tie                    0.40%

Mean cards consumed:
  Textbook             165.674
  Maximum Aggression   177.840

HOLDOUT
Average ending bankroll:
  Textbook             99.972
  Maximum Aggression   78.895

Positive sessions:
  Textbook             47.42%
  Maximum Aggression    2.42%

Paired winner:
  Textbook             95.86%
  Maximum Aggression    3.86%
  Tie                    0.28%

Mean cards consumed:
  Textbook             165.647
  Maximum Aggression   177.749

PLAIN-ENGLISH INTERPRETATION
The independent Primary and Holdout cohorts tell essentially the same story. The frozen Maximum Aggression player finishes far below the Textbook player on average and only rarely finishes the 30-hand session above its starting bankroll.

That is useful for the new programme because the baseline is not ambiguous: there is substantial behavioural damage available to investigate.

Phase 3.01's important contribution is therefore not a new strategy. It is the measurement framework that allows us to locate repeated damage at the level of visible decision states and then alter one policy family at a time.

IMPORTANT FOLLOW-UP DISCOVERED AFTER 3.01
The first state rankings appeared to contain hard-20 DOUBLE states. That raised an important question: were initial combinations such as K+J or K+Q incorrectly being doubled instead of split?

Rather than begin optimisation with that ambiguity unresolved, Phase 3.02 became a quality-control audit.

The historical/code audit subsequently confirmed that 10, J, Q and K are all stored as numerical value 10. Initial two-card combinations such as K+J, K+Q, J+Q and 10+K are therefore split-eligible and are routed to SPLIT before Maximum Aggression reaches its DOUBLE rule.

The hard-20 DOUBLE labels in the state evidence can arise from split-child states under the one-split architecture; they do not demonstrate that an unsplit initial K+J-type hand was doubled.

WHAT PHASE 3.01 PROVES
- The Phase 2.02 finite-shoe baseline reproduces under the new instrumentation.
- Maximum Aggression remains a strongly separated extreme baseline.
- State-level instrumentation can support controlled policy optimisation.
- Primary and holdout evidence can be compared before a rule is changed.

WHAT IT DOES NOT PROVE
- It does not identify a proven improved policy.
- It does not prove that any single state family should yet be removed or changed.
- It does not use card counting or live shoe-composition information.
- It does not change the historical Maximum Aggression definition.

WHY PHASE 3.02 FOLLOWED
Before changing Maximum Aggression, the ten-value split-precedence question had to be resolved. Phase 3.02 therefore became a verification gateway rather than the first optimisation intervention.

The first genuine one-rule stabilisation intervention is consequently reserved for Phase 3.03.
