DETERMINISTIC BLACKJACK — MAXIMUM AGGRESSION POLICY OPTIMISATION
RUNNING RESEARCH GUIDE

Programme start: Phase 3.01
Current documented phase: Phase 3.32

======================================================================
1. WHY THIS NEW PROGRAMME EXISTS
======================================================================
The completed blackjack research established a broad evidence base comparing textbook and non-textbook behaviour, including a deterministic finite six-deck shoe and an intentionally extreme Maximum Aggression player.

The Phase 3.xx programme asks a narrower new question:

Can Maximum Aggression be stabilised into a selective, evidence-driven policy that improves its probability of finishing a fixed 30-hand session with a higher bankroll, without using card counting, hidden information or future-card knowledge?

The immediate target is not to maximise a 2x, 3x, 5x or 10x bankroll. First, the programme must determine whether indiscriminate aggression can be made materially more stable. Larger bankroll targets belong to later work only after the policy mechanism is understood and replicated.

======================================================================
2. EVIDENCE BASIS AND HISTORICAL BOUNDARY
======================================================================
PRIMARY DESIGN AUTHORITY
The optimisation programme is driven primarily by the deterministic finite six-deck evidence from Phase 2.xx and by new Phase 3.xx primary/holdout datasets.

SECONDARY HISTORICAL EVIDENCE
Earlier non-depleting/infinite-generator behaviour-profile experiments remain useful for historical comparison, hypothesis generation and corroboration. They do not directly define a new optimisation rule unless the finite-shoe evidence independently supports the same conclusion.

PROGRESSIVE EVIDENCE RULE
As Phase 3.xx develops, newly generated controlled optimisation datasets become the strongest evidence for subsequent policy changes.

======================================================================
3. FROZEN STARTING ARCHITECTURE
======================================================================
The programme initially preserves the established finite-shoe architecture so that old and new evidence remain comparable:

- Six decks / 312-card shoe.
- Deterministic session shuffle.
- 30 hands per session.
- 100-unit starting bankroll.
- 1-unit base wager.
- Existing dealer rules and blackjack payout.
- Existing split limit and double rules.
- No in-session reshuffle at the optimisation-entry stage.
- Textbook strategy remains the control/reference.
- Frozen Maximum Aggression remains the extreme baseline.

Thirty hands is retained initially as the established experimental horizon. Alternative session lengths may later be examined as a separate sensitivity study; they should not be changed at the same time as the first policy rules.

A realistic interval/penetration reshuffle is also reserved for a later robustness stage. It should not be introduced simultaneously with initial policy optimisation because doing so would make attribution harder.

======================================================================
4. INFORMATION BOUNDARY FOR THE NEW POLICY
======================================================================
The optimiser may eventually use information legitimately visible at the current decision point, including:

- player total and hand composition;
- hard/soft status;
- pair status;
- dealer upcard;
- legal actions;
- previously validated experimental risk/outcome trends associated with those visible states.

The initial programme does NOT give the policy a running count, previous-card-history feature, remaining-rank inventory, hidden-card information or knowledge of future cards.

Historical experimental trends may inform a rule before a new run begins. That is different from adapting a decision during the run because particular ranks have already left the current shoe.

======================================================================
5. REFERENCE POLICIES
======================================================================
TEXTBOOK
The established basic-strategy implementation remains the control/reference policy.

MAXIMUM AGGRESSION
Maximum Aggression is retained as the deliberately extreme boundary condition. Its action precedence is important:

1. If the initial two-card state is split-eligible, SPLIT first.
2. Otherwise, DOUBLE on the first legal decision under the frozen aggressive rule.
3. Continue according to the established Maximum Aggression implementation.

Because the simulator stores 10, J, Q and K as numerical value 10, all initial two-card combinations of these ten-valued ranks are treated as 10+10 for split eligibility. Thus K+J, K+Q, J+Q, 10+K and equivalent combinations are routed to SPLIT before DOUBLE.

======================================================================
6. PHASE 3.01 — FROZEN BASELINE AND OPTIMISATION ENTRY POINT
======================================================================
PURPOSE
Phase 3.01 entered the optimisation programme without changing Maximum Aggression. Its job was to preserve the finite six-deck baseline while adding state-level instrumentation capable of identifying where aggressive decisions cause repeated bankroll damage.

ARCHIVAL STATUS
The original Phase 3.01 experiment package has now been preserved as the official archival package and supplemented with a retrospective layman results explanation. This documentation addition does not change the Phase 3.01 Java code, datasets, output or experimental results.

WHAT WAS ADDED
The evidence record was expanded to include visible starting state, hard/soft status, dealer upcard, split-child status, textbook first action for the same visible state, aggressive action, stake, cards consumed, third-card behaviour, hand outcome and bankroll contribution.

WHAT WAS NOT CHANGED
The Maximum Aggression policy itself was not optimised. The six-deck architecture, 30-hand horizon, bankroll start and textbook control remained frozen.

WHY IT MATTERED
The instrumentation made it possible to move away from vague labels such as 'more aggressive' and instead identify adequately sampled visible-state families whose losses replicate in both primary and holdout evidence.

IMPORTANT FOLLOW-UP
Review of the initial state labels raised a question about hard-20 doubles and whether ten-valued combinations such as K+J were being split first. This required an explicit historical/code audit before optimisation could safely begin.

DOCUMENTED DESIGN EVOLUTION
The original Phase 3.01 next-design note anticipated that Phase 3.02 would make the first one-rule intervention. The hard-20 / ten-value split question emerged before that intervention was accepted, so Phase 3.02 was correctly reassigned as a verification gateway. The first genuine stabilisation intervention therefore moves to Phase 3.03.

======================================================================
7. PHASE 3.02 — TEN-VALUE SPLIT PRECEDENCE AUDIT
======================================================================
PURPOSE
Phase 3.02 resolved the ten-value split question before making any policy change.

AUDIT RESULT
PASS. The simulator represents 10, J, Q and K as integer value 10. Maximum Aggression evaluates split eligibility before the aggressive double rule. Therefore all initial two-card ten-value combinations are split-eligible and are routed to SPLIT first.

Examples:
10+J -> stored 10+10 -> SPLIT
K+Q  -> stored 10+10 -> SPLIT
J+K  -> stored 10+10 -> SPLIT
Q+Q  -> stored 10+10 -> SPLIT

This behaviour is consistent with the historical baseline. Phase 3.02 clarifies the interpretation rather than changing previous evidence.

BASELINE REPLICATION
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.

STATE-LEVEL CONTEXT
Primary resolved aggressive hands: 171,838; mean net per resolved hand -0.6117; bust rate 35.76%.
Holdout resolved aggressive hands: 171,789; mean net per resolved hand -0.6143; bust rate 35.72%.

The state ranking contains HARD 20 DOUBLE states marked [split-child]. These are not unsplit initial K+J-type hands. Under the one-split limit, a child created by a split may receive another ten-valued card and subsequently reach the aggressive DOUBLE rule.

LAYMAN INTERPRETATION
Phase 3.02 is a quality-control experiment. Before trying to make Maximum Aggression better, it confirms exactly what the aggressive player is already doing. The very poor bankroll result remains replicated in independent primary and holdout cohorts, while the split-precedence ambiguity is removed.

WHAT PHASE 3.02 DOES NOT CLAIM
It does not identify a proven improved policy. It does not use card counting or live shoe-composition information. It does not change the historical Maximum Aggression definition.

NEXT DESIGN — PHASE 3.03
Phase 3.03 should make the first genuine stabilisation change. Only ONE replicated visible-state policy family should be altered, chosen from adequately sampled finite-shoe evidence that shows the same damaging direction in both primary and holdout cohorts.

The purpose is attribution: if bankroll behaviour changes, we should be able to say which policy intervention caused it.

======================================================================
8. DOCUMENTATION AND PUBLICATION WORKFLOW
======================================================================
Each Phase 3.xx experiment should preserve four levels of evidence:

1. Java code — exact implementation.
2. Raw/detailed results and CSV evidence — audit trail.
3. Individual experiment results TXT — concise numerical results plus plain-English explanation.
4. This cumulative Research Guide — the connected scientific narrative across all experiments.

Publication plates are selective. A plate should be created when an experiment establishes an important baseline, milestone, conceptual transition or finding that benefits from visual explanation. Plates should explain why the programme has reached that point, which evidence is being used, what remains frozen, what is being tested and what the evidence means.


======================================================================
9. PHASE 3.03 — FIRST CONTROLLED STABILISATION INTERVENTION
======================================================================
PURPOSE
Phase 3.03 makes the first genuine policy change in the optimisation programme. It changes one replicated visible-state family only: a split-child HARD 20 now STANDS at its first decision instead of taking Maximum Aggression's automatic DOUBLE.

EVIDENCE FOR SELECTION
Phase 3.02 Primary and Holdout rankings independently placed split-child HARD 20 states at the top of the adequately sampled damage table. Across dealer upcards, the automatic DOUBLE produced approximately -1.65 to -1.77 mean units per resolved hand with roughly 91-93% bust rates; Textbook would STAND.

WHAT REMAINED FROZEN
The six-deck finite shoe, deterministic session-start shuffle, 30-hand horizon, 100-unit start, 1-unit wager, dealer/payout/split architecture, no in-session reshuffle, Textbook control and information boundary all remain unchanged. Initial ten-value split precedence remains unchanged. Every other Maximum Aggression first-decision state still follows the frozen aggressive rule.

PRIMARY RESULTS
Average ending bankroll — Textbook: 99.956; Selective Aggression: 82.767.
Positive sessions — Textbook: 48.14%; Selective Aggression: 5.48%.
Paired winner — Textbook: 92.04%; Selective Aggression: 7.54%; tie: 0.42%.
Mean cards consumed — Textbook: 165.674; Selective Aggression: 176.838.

HOLDOUT RESULTS
Average ending bankroll — Textbook: 99.972; Selective Aggression: 82.748.
Positive sessions — Textbook: 47.42%; Selective Aggression: 5.10%.
Paired winner — Textbook: 92.80%; Selective Aggression: 6.64%; tie: 0.56%.
Mean cards consumed — Textbook: 165.647; Selective Aggression: 176.801.

COMPARISON WITH FROZEN MAXIMUM AGGRESSION
Phase 3.02 Maximum Aggression averaged 78.978 Primary / 78.895 Holdout and produced 2.48% / 2.42% positive sessions. Phase 3.03 therefore recovers about 3.8 average bankroll units in both cohorts and raises positive-session frequency to 5.48% / 5.10%. The direction and approximate magnitude replicate independently.

MULTIPLIER MONITORING
From Phase 3.03 onward, ending bankroll thresholds >=200 (2x), >=300 (3x), >=500 (5x) and >=1000 (10x) are recorded as observational outcomes. All were 0.00% in both Phase 3.03 cohorts. These thresholds do not influence the policy at this stage.

LAYMAN INTERPRETATION
The first surgical correction worked in the intended direction. Maximum Aggression remains a poor player overall, but simply refusing to double a split-child hard 20 materially reduces damage. This is the first evidence that the extreme baseline can be stabilised incrementally using visible-state evidence rather than card counting or live shoe history.

LIMITATION
This experiment validates only one intervention. It does not establish the final selective-aggression policy and does not justify changing several neighbouring states together.

NEXT DESIGN — PHASE 3.04
Carry forward the Phase 3.03 HARD 20 split-child STAND rule. The remaining replicated ranking is now dominated by HARD 19 states. Phase 3.04 should select and alter only one next adequately sampled HARD 19 family, preserving one-rule-at-a-time attribution and all frozen architecture.

======================================================================
10. PHASE 3.04 — SECOND CONTROLLED STABILISATION INTERVENTION
======================================================================
PURPOSE
Phase 3.04 carries forward the Phase 3.03 split-child HARD 20 STAND rule and adds one further visible-state policy family only: HARD 19 now STANDS at its first decision instead of taking Maximum Aggression's automatic DOUBLE. The new HARD 19 rule applies to ordinary and split-child hands.

WHY HARD 19
After the Phase 3.03 intervention, HARD 19 states dominated the remaining replicated damage ranking across Primary and Holdout evidence. The family was strongly negative across dealer upcards, with approximately 80% bust rates under the aggressive DOUBLE, while Textbook uniformly STANDS on HARD 19.

WHAT REMAINED FROZEN
The six-deck finite shoe, deterministic session-start shuffle, 30-hand horizon, 100-unit start, 1-unit base wager, dealer/payout/split architecture, no surrender, no in-session reshuffle, Textbook control and information boundary remain unchanged. The Phase 3.03 split-child HARD 20 rule remains active. Every other Maximum Aggression state remains unchanged.

PRIMARY RESULTS
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 RESULTS
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.

COMPARISON WITH PHASE 3.03
Phase 3.03 Selective Aggression averaged 82.767 Primary / 82.748 Holdout with 5.48% / 5.10% positive sessions. Phase 3.04 therefore adds +3.52 / +3.28 average bankroll units and +3.92 / +3.66 percentage points of positive-session frequency. The direction again replicates independently.

CUMULATIVE POLICY LAYERS
1. Phase 3.03 — after an initial split, if a child hand subsequently forms HARD 20, STAND rather than DOUBLE.
2. Phase 3.04 — if a first-decision hand is HARD 19, STAND rather than DOUBLE, whether ordinary or split-child.

MULTIPLIER MONITORING
Ending-bankroll thresholds >=200 (2x), >=300 (3x), >=500 (5x) and >=1000 (10x) remain observational only. All were 0.00% in both Phase 3.04 cohorts.

LAYMAN INTERPRETATION
A second precisely targeted correction again improves the Maximum Aggression bankroll without replacing the whole policy with Textbook strategy. The player remains substantially worse than Textbook, but the evidence now shows a replicated progression: removing two specific forms of indiscriminate doubling materially reduces damage.

NEXT DESIGN — PHASE 3.05
Carry forward both validated rules. HARD 18 now dominates the remaining replicated damage table. Phase 3.05 should investigate one coherent HARD 18 family only, with all architecture and information controls frozen.


======================================================================
11. PHASE 3.05 — THIRD CONTROLLED STABILISATION INTERVENTION
======================================================================
PURPOSE
Phase 3.05 carries forward the validated split-child HARD 20 STAND rule and the HARD 19 STAND family. It adds one further family only: split-child HARD 18 now STANDS instead of taking Maximum Aggression's automatic DOUBLE. Ordinary HARD 18 remains unchanged.

WHY SPLIT-CHILD HARD 18
Phase 3.04 showed HARD 18 dominating the remaining replicated damage evidence. The publication next-design decision deliberately narrowed 3.05 to split-child HARD 18 so the third intervention remains surgical and directly extends the post-split protection already established at HARD 20.

WHAT REMAINED FROZEN
Six decks, deterministic session-start shuffle, 30-hand horizon, 100-unit start, 1-unit base wager, dealer/payout/split architecture, no surrender, no in-session reshuffle, Textbook control and the information boundary remain unchanged. No card counting, card history, remaining-shoe composition, hidden-card or future-card information is introduced.

PRIMARY RESULTS
Average ending bankroll — Textbook: 99.956; Selective Aggression: 86.922.
Positive sessions — Textbook: 48.14%; Selective Aggression: 10.18%.
Paired winner — Textbook: 86.34%; Selective Aggression: 12.50%; tie: 1.16%.
Mean cards consumed — Textbook: 165.674; Selective Aggression: 175.828.

HOLDOUT RESULTS
Average ending bankroll — Textbook: 99.972; Selective Aggression: 86.706.
Positive sessions — Textbook: 47.42%; Selective Aggression: 9.80%.
Paired winner — Textbook: 87.34%; Selective Aggression: 11.68%; tie: 0.98%.
Mean cards consumed — Textbook: 165.647; Selective Aggression: 175.694.

COMPARISON WITH PHASE 3.04
Phase 3.04 averaged 86.287 Primary / 86.023 Holdout with 9.40% / 8.76% positive sessions. Phase 3.05 therefore adds +0.635 / +0.683 average bankroll units and +0.78 / +1.04 percentage points of positive-session frequency. The improvement is modest but independently directionally replicated.

CUMULATIVE POLICY LAYERS
1. Phase 3.03 — split-child HARD 20 -> STAND.
2. Phase 3.04 — HARD 19 -> STAND, ordinary or split-child.
3. Phase 3.05 — split-child HARD 18 -> STAND.
All other Maximum Aggression behaviour remains frozen.

MULTIPLIER MONITORING
Ending-bankroll thresholds >=200 (2x), >=300 (3x), >=500 (5x) and >=1000 (10x) remain observational only. All were 0.00% in both Phase 3.05 cohorts.

PEAK-BANKROLL INSTRUMENTATION
Phase 3.05 introduces direct bankroll tracking after every hand. Selective Aggression reached a cohort-wide maximum of 127.0 units at Hand 30 in Primary and 129.0 units at Hand 18 in Holdout.

LAYMAN INTERPRETATION
Protecting split-child HARD 18 helps again, but much less dramatically than protecting HARD 20 and HARD 19. This is scientifically useful because it begins to show diminishing marginal benefit rather than assuming every additional restriction will produce a large gain.

NEXT DESIGN — PHASE 3.06
Ordinary HARD 18 remains unchanged and continues to dominate the replicated remaining-damage evidence. Phase 3.06 should test extending the HARD 18 STAND protection to ordinary first-decision HARD 18 only, while preserving every other control and the accumulated policy layers.


======================================================================
12. PHASE 3.06 — FOURTH CONTROLLED STABILISATION INTERVENTION
======================================================================
PURPOSE
Phase 3.06 carries forward split-child HARD 20 STAND, HARD 19 STAND and split-child HARD 18 STAND. It adds one further visible-state family only: ordinary HARD 18 now STANDS instead of taking Maximum Aggression's automatic DOUBLE.

WHY ORDINARY HARD 18
Phase 3.05 deliberately left ordinary HARD 18 unchanged. The remaining replicated evidence continued to show it prominently, allowing its contribution to be tested separately rather than bundled into the split-child intervention.

WHAT REMAINED FROZEN
The six-deck finite shoe, deterministic session-start shuffle, 30-hand horizon, 100-unit start, 1-unit base wager, dealer/payout/split architecture, no surrender, no in-session reshuffle, Textbook control and information boundary remain unchanged. Peak-bankroll instrumentation is observational only.

PRIMARY RESULTS
Average ending bankroll — Textbook: 99.956; Selective Aggression: 88.684.
Positive sessions — Textbook: 48.14%; Selective Aggression: 12.70%.
Paired winner — Textbook: 83.96%; Selective Aggression: 14.90%; tie: 1.14%.
Mean cards consumed — Textbook: 165.674; Selective Aggression: 175.528.
Maximum Selective-Aggression bankroll observed: 135.0 at Hand 26.

HOLDOUT RESULTS
Average ending bankroll — Textbook: 99.972; Selective Aggression: 88.461.
Positive sessions — Textbook: 47.42%; Selective Aggression: 12.18%.
Paired winner — Textbook: 84.84%; Selective Aggression: 14.20%; tie: 0.96%.
Mean cards consumed — Textbook: 165.647; Selective Aggression: 175.355.
Maximum Selective-Aggression bankroll observed: 132.0 at Hand 30.

COMPARISON WITH PHASE 3.05
Phase 3.05 Selective Aggression averaged 86.922 Primary / 86.706 Holdout with 10.18% / 9.80% positive sessions. Phase 3.06 therefore adds +1.762 / +1.755 average bankroll units and +2.52 / +2.38 percentage points of positive-session frequency.

CUMULATIVE POLICY LAYERS
1. Phase 3.03 — split-child HARD 20 -> STAND.
2. Phase 3.04 — HARD 19 -> STAND, ordinary or split-child.
3. Phase 3.05 — split-child HARD 18 -> STAND.
4. Phase 3.06 — ordinary HARD 18 -> STAND.
All other Maximum Aggression behaviour remains frozen.

MULTIPLIER MONITORING
Ending-bankroll thresholds >=200 (2x), >=300 (3x), >=500 (5x) and >=1000 (10x) remain observational only. All were 0.00% in both Phase 3.06 cohorts.

PEAK-BANKROLL OBSERVATION
The direct per-hand measurement records 135.0 units at Hand 26 in Primary and 132.0 units at Hand 30 in Holdout. These maxima are useful observations for later examination of the 30-hand horizon and peak-to-end retention, but are not used by the policy and are not treated as evidence of a repeatable route by themselves.

LAYMAN INTERPRETATION
Extending HARD 18 protection to ordinary hands produces another clear replicated improvement. The gain is larger than the split-child-only HARD 18 change in Phase 3.05, while remaining smaller than the early HARD 19 correction. This supports continuing controlled, one-family-at-a-time stabilisation.

NEXT DESIGN — PHASE 3.07
The leading remaining replicated damage row after Phase 3.06 is:
1 | HARD 16 vs 11 [split-child]        |   162 / -1.309 |   174 / -1.207 |  63.58% /  60.92% | HIT
Phase 3.07 should select one coherent family from the replicated Primary/Holdout evidence and preserve all other controls.


======================================================================
13. PHASE 3.07 — FIFTH CONTROLLED STABILISATION INTERVENTION
======================================================================
PURPOSE
Phase 3.07 carries forward all four validated protections from 3.03–3.06 and changes one exact visible state only: split-child HARD 16 versus dealer Ace now HITS, matching the Textbook first action, rather than taking Maximum Aggression's automatic DOUBLE.

WHY THIS STATE
It was the leading remaining replicated damage row after 3.06: Primary n=162 / mean net per hand -1.309; Holdout n=174 / -1.207. The direction and severity replicated independently.

WHAT REMAINED FROZEN
Six-deck finite shoe, deterministic session-start shuffle, 30-hand horizon, 100-unit start, flat 1-unit wager, S17, 3:2 blackjack, one split, DAS otherwise permitted, no surrender, no in-session reshuffle, Textbook control, information boundary, multiplier monitoring and peak-bankroll instrumentation.

RESULTS
Primary Selective Aggression average ending bankroll: 88.705, versus 88.684 in 3.06 (+0.021).
Holdout: 88.482, versus 88.461 (+0.021).
Positive sessions: 12.72% Primary and 12.18% Holdout.
No session finished at 2x, 3x, 5x or 10x.
Maximum observed Selective-Aggression bankroll remains 135.0 at Hand 26 Primary and 132.0 at Hand 30 Holdout.

INTERPRETATION
The correction works directionally but its population-level effect is tiny. Phase 3.07 therefore demonstrates an important distinction between per-encounter damage and overall optimisation leverage. A severely damaging state can be too uncommon to move the aggregate bankroll distribution materially.

CUMULATIVE POLICY LAYERS
1. 3.03 — split-child HARD 20 -> STAND.
2. 3.04 — HARD 19 -> STAND.
3. 3.05 — split-child HARD 18 -> STAND.
4. 3.06 — ordinary HARD 18 -> STAND.
5. 3.07 — split-child HARD 16 vs Ace -> HIT.
All other Maximum Aggression behaviour remains frozen.

NEXT DESIGN — PHASE 3.08
The remaining evidence is now heavily populated by HARD 17 states. The leading individual row is split-child HARD 17 vs dealer 6 (Textbook STAND), but multiple ordinary and split-child HARD 17 rows are also prominent. Phase 3.08 should decide from the replicated evidence whether a coherent HARD 17 family can be justified or whether another dealer-specific intervention is required. Preserve one-change-at-a-time attribution.


======================================================================
14. PHASE 3.08 — SIXTH CONTROLLED STABILISATION INTERVENTION
======================================================================
PURPOSE
Phase 3.08 carries forward all validated 3.03–3.07 protections and changes one coherent decision family only: any first-decision HARD 17 now STANDS rather than taking Maximum Aggression's automatic DOUBLE.

WHY A HARD 17 FAMILY
Phase 3.07 demonstrated that severe per-encounter damage can have very little aggregate leverage when exposure is low. After 3.07, HARD 17 states recurred prominently across multiple dealer upcards and in ordinary and split-child contexts. The corresponding Textbook first action was consistently STAND. The experiment therefore tests one coherent decision principle rather than another isolated row.

WHAT REMAINED FROZEN
Six-deck finite shoe, deterministic session-start shuffle, 30-hand horizon, 100-unit start, 1-unit base wager, S17, 3:2 blackjack, one split, DAS otherwise permitted, no surrender, no in-session reshuffle, Textbook control and the strict information boundary. Multiplier and peak-bankroll measurements remain observational only.

PRIMARY RESULTS
Average ending bankroll — Textbook: 99.956; Selective Aggression: 90.594.
Positive sessions — Textbook: 48.14%; Selective Aggression: 16.14%.
Paired winner — Textbook: 80.26%; Selective Aggression: 18.42%; tie: 1.32%.
Mean cards consumed — Textbook: 165.674; Selective Aggression: 174.385.
Maximum Selective-Aggression bankroll observed: 130.0 at Hand 30.

HOLDOUT RESULTS
Average ending bankroll — Textbook: 99.972; Selective Aggression: 90.478.
Positive sessions — Textbook: 47.42%; Selective Aggression: 16.14%.
Paired winner — Textbook: 79.90%; Selective Aggression: 18.42%; tie: 1.68%.
Mean cards consumed — Textbook: 165.647; Selective Aggression: 174.221.
Maximum Selective-Aggression bankroll observed: 132.0 at Hand 30.

COMPARISON WITH PHASE 3.07
Phase 3.07 averaged 88.705 Primary / 88.482 Holdout with 12.72% / 12.18% positive sessions. Phase 3.08 therefore adds +1.889 / +1.996 average bankroll units and +3.42 / +3.96 percentage points of positive-session frequency.

CUMULATIVE POLICY LAYERS
1. 3.03 — split-child HARD 20 -> STAND.
2. 3.04 — HARD 19 -> STAND.
3. 3.05 — split-child HARD 18 -> STAND.
4. 3.06 — ordinary HARD 18 -> STAND.
5. 3.07 — split-child HARD 16 vs Ace -> HIT.
6. 3.08 — HARD 17 -> STAND, ordinary or split-child.
All other Maximum Aggression behaviour remains frozen.

MULTIPLIER AND PEAK OBSERVATION
No session finished at 2x, 3x, 5x or 10x. The Selective-Aggression peak was 130.0 at Hand 30 Primary and 132.0 at Hand 30 Holdout. The lower Primary extreme compared with 3.07 occurs alongside materially better average and positive-session outcomes, reinforcing that a single maximum peak and population-level optimisation are different measurements.

INTERPRETATION
Phase 3.08 strongly supports incorporating exposure/frequency into candidate selection. A coherent repeatedly encountered family can provide much greater aggregate leverage than a rarer state with worse per-encounter damage.

NEXT DESIGN — PHASE 3.09
The remaining replicated evidence is heavily populated by HARD 16 states. Unlike HARD 17, their correct Textbook action is dealer-dependent: recorded HARD 16 states include both HIT and STAND. Phase 3.09 should therefore investigate HARD 16 as a structured dealer-dependent family, mapping qualifying states to their corresponding Textbook action rather than imposing one blanket rule.


======================================================================
15. PHASE 3.09 — SEVENTH CONTROLLED STABILISATION INTERVENTION
======================================================================
PURPOSE
Phase 3.09 carries forward every validated 3.03–3.08 layer and changes one structured family only: first-decision HARD 16 follows its corresponding dealer-dependent Textbook action instead of Maximum Aggression's automatic DOUBLE.

DECISION MAPPING
HARD 16 vs dealer 2–6 -> STAND.
HARD 16 vs dealer 7–Ace -> HIT.
This preserves the state-dependent structure already present in the frozen Textbook control rather than imposing an incorrect blanket action.

WHY HARD 16
After 3.08, HARD 16 appeared repeatedly and with substantial exposure in the replicated remaining-damage evidence. Unlike HARD 17, the correct action varied by dealer upcard. Phase 3.09 therefore tests a coherent mapped family rather than a single universal action.

PRIMARY RESULTS
Average ending bankroll — Textbook: 99.956; Selective Aggression: 91.986.
Positive sessions — Textbook: 48.14%; Selective Aggression: 19.56%.
Paired winner — Textbook: 77.12%; Selective Aggression: 21.40%; tie: 1.48%.
Mean cards consumed — Textbook: 165.674; Selective Aggression: 174.145.
Maximum Selective-Aggression bankroll: 130.0 at Hand 30.

HOLDOUT RESULTS
Average ending bankroll — Textbook: 99.972; Selective Aggression: 91.741.
Positive sessions — Textbook: 47.42%; Selective Aggression: 18.66%.
Paired winner — Textbook: 77.54%; Selective Aggression: 20.92%; tie: 1.54%.
Mean cards consumed — Textbook: 165.647; Selective Aggression: 173.973.
Maximum Selective-Aggression bankroll: 130.0 at Hand 18.

COMPARISON WITH PHASE 3.08
Average bankroll improves +1.392 Primary and +1.263 Holdout.
Positive-session frequency improves +3.42 pp Primary and +2.52 pp Holdout.
No session finishes at 2x, 3x, 5x or 10x.

CUMULATIVE POLICY LAYERS
1. 3.03 — split-child HARD 20 -> STAND.
2. 3.04 — HARD 19 -> STAND.
3. 3.05 — split-child HARD 18 -> STAND.
4. 3.06 — ordinary HARD 18 -> STAND.
5. 3.07 — split-child HARD 16 vs Ace -> HIT.
6. 3.08 — HARD 17 -> STAND.
7. 3.09 — HARD 16 -> dealer-dependent Textbook action.
The 3.09 family naturally subsumes the previously validated 3.07 HARD 16 vs Ace state while remaining consistent with it.

INTERPRETATION
A coherent intervention family need not prescribe one identical action. It can preserve a deterministic mapping from the visible player state and dealer upcard to the established control action. This allows broader leverage without discarding experimental attribution or introducing card-history information.

NEXT DESIGN — PHASE 3.10
The remaining evidence shifts toward HARD 15, HARD 14 and low-total HIT states. Candidate selection should now explicitly combine replicated per-encounter damage with exposure. HARD 15 is a strong family for investigation because several rows remain prominent and HARD 15 vs dealer 10 alone has approximately 3,000 observations per cohort. Phase 3.10 should determine the narrowest coherent HARD 15 intervention justified by the replicated evidence.


======================================================================
16. PHASE 3.10 — EIGHTH CONTROLLED STABILISATION INTERVENTION
======================================================================
PURPOSE
Phase 3.10 carries forward all validated 3.03–3.09 layers and changes one structured family only: first-decision HARD 15 follows its corresponding dealer-dependent Textbook action instead of Maximum Aggression's automatic DOUBLE.

DECISION MAPPING
HARD 15 vs dealer 2–6 -> STAND.
HARD 15 vs dealer 7–Ace -> HIT.

WHY HARD 15
After 3.09, HARD 15 remained prominent in the replicated damage evidence and some states had very high exposure. HARD 15 vs dealer 10 alone occurred approximately 3,000 times per cohort. Phase 3.10 therefore applies the lesson from 3.07: intervention priority should reflect both per-encounter damage and programme-level exposure.

PRIMARY RESULTS
Average ending bankroll — Textbook: 99.956; Selective Aggression: 93.251.
Positive sessions — Textbook: 48.14%; Selective Aggression: 22.54%.
Paired winner — Textbook: 73.72%; Selective Aggression: 24.62%; tie: 1.66%.
Mean cards consumed — Textbook: 165.674; Selective Aggression: 173.737.
Maximum Selective-Aggression bankroll: 130.0 at Hand 30.

HOLDOUT RESULTS
Average ending bankroll — Textbook: 99.972; Selective Aggression: 93.193.
Positive sessions — Textbook: 47.42%; Selective Aggression: 22.02%.
Paired winner — Textbook: 74.56%; Selective Aggression: 23.80%; tie: 1.64%.
Mean cards consumed — Textbook: 165.647; Selective Aggression: 173.617.
Maximum Selective-Aggression bankroll: 131.0 at Hand 30.

COMPARISON WITH PHASE 3.09
Average bankroll improves +1.265 Primary and +1.452 Holdout.
Positive-session frequency improves +2.98 pp Primary and +3.36 pp Holdout.
No session finishes at 2x, 3x, 5x or 10x.

CUMULATIVE POLICY LAYERS
1. 3.03 — split-child HARD 20 -> STAND.
2. 3.04 — HARD 19 -> STAND.
3. 3.05 — split-child HARD 18 -> STAND.
4. 3.06 — ordinary HARD 18 -> STAND.
5. 3.07 — split-child HARD 16 vs Ace -> HIT.
6. 3.08 — HARD 17 -> STAND.
7. 3.09 — HARD 16 -> dealer-dependent Textbook action.
8. 3.10 — HARD 15 -> dealer-dependent Textbook action.
All other Maximum Aggression behaviour remains frozen.

INTERPRETATION
Phase 3.10 continues the replicated improvement and reinforces the value of high-exposure coherent families. The optimisation programme is now selectively importing only those Textbook decisions justified by finite-shoe Primary/Holdout evidence.

EMERGING MECHANISM CHANGE
The remaining table no longer consists only of high-total bust-heavy states. Low totals such as HARD 5, 6 and 7 against strong dealer upcards now appear prominently with 0% immediate bust rates. Their poor outcome under automatic DOUBLE therefore represents a different damage mechanism from the earlier HARD 20–15 progression.

NEXT DESIGN — PHASE 3.11
Phase 3.11 should distinguish remaining mid-total HARD 14-type states from low-total automatic-DOUBLE states. Candidate selection should combine replicated damage, exposure, Textbook-action coherence and damage mechanism rather than mechanically descending totals.


======================================================================
17. PHASE 3.11 — NINTH CONTROLLED STABILISATION INTERVENTION
======================================================================
PURPOSE
Phase 3.11 carries forward every validated 3.03–3.10 layer and changes one structured family only: first-decision HARD 14 follows its corresponding dealer-dependent Textbook action instead of Maximum Aggression's automatic DOUBLE.

DECISION MAPPING
HARD 14 vs dealer 2–6 -> STAND.
HARD 14 vs dealer 7–Ace -> HIT.

WHY HARD 14
Phase 3.10 separated the remaining evidence into mid-total bust-exposed states and low-total zero-immediate-bust states. HARD 14 was selected as the next coherent repeatedly exposed mid-total family, deliberately leaving the distinct low-total mechanism untouched.

PRIMARY RESULTS
Average ending bankroll — Textbook: 99.956; Selective Aggression: 94.497.
Positive sessions — Textbook: 48.14%; Selective Aggression: 26.20%.
Paired winner — Textbook: 70.02%; Selective Aggression: 28.10%; tie: 1.88%.
Mean cards consumed — Textbook: 165.674; Selective Aggression: 173.413.
Maximum Selective-Aggression bankroll: 126.0 at Hand 30.

HOLDOUT RESULTS
Average ending bankroll — Textbook: 99.972; Selective Aggression: 94.466.
Positive sessions — Textbook: 47.42%; Selective Aggression: 25.70%.
Paired winner — Textbook: 70.78%; Selective Aggression: 27.44%; tie: 1.78%.
Mean cards consumed — Textbook: 165.647; Selective Aggression: 173.296.
Maximum Selective-Aggression bankroll: 128.5 at Hand 30.

COMPARISON WITH PHASE 3.10
Average bankroll improves +1.246 Primary and +1.273 Holdout.
Positive-session frequency improves +3.66 pp Primary and +3.68 pp Holdout.
No session finishes at 2x, 3x, 5x or 10x.

CUMULATIVE POLICY LAYERS
1. 3.03 — split-child HARD 20 -> STAND.
2. 3.04 — HARD 19 -> STAND.
3. 3.05 — split-child HARD 18 -> STAND.
4. 3.06 — ordinary HARD 18 -> STAND.
5. 3.07 — split-child HARD 16 vs Ace -> HIT.
6. 3.08 — HARD 17 -> STAND.
7. 3.09 — HARD 16 -> dealer-dependent Textbook action.
8. 3.10 — HARD 15 -> dealer-dependent Textbook action.
9. 3.11 — HARD 14 -> dealer-dependent Textbook action.
All other Maximum Aggression behaviour remains frozen.

INTERPRETATION
The mid-total sequence continues to produce substantial replicated recovery. The Selective-Aggression average is now approximately 94.5 units in both cohorts, considerably closer to the approximately 100-unit Textbook control while retaining aggressive behaviour outside the evidence-selected intervention layers.

SECOND DAMAGE MECHANISM NOW DOMINANT
The remaining replicated ranking is increasingly populated by HARD 5, 6 and 7 states with 0% immediate bust rates and Textbook HIT actions. Their negative outcome therefore cannot be explained by immediate busting. The aggressive policy is doubling the wager and accepting the forced one-card double-down sequence where Textbook would retain HIT flexibility.

NEXT DESIGN — PHASE 3.12
Phase 3.12 should define and test the narrowest coherent low-total automatic-DOUBLE family supported by replicated damage, exposure, Textbook-action consistency and the shared zero-immediate-bust mechanism. Do not blanket-convert all low totals without establishing the boundary.


======================================================================
18. PHASE 3.12 — LOW-TOTAL FORCED-ONE-CARD MECHANISM
======================================================================
PURPOSE
Phase 3.12 carries forward every validated 3.03–3.11 policy layer and tests one bounded low-total family only: HARD 5–7 versus dealer 7–Ace now HIT instead of Maximum Aggression's automatic DOUBLE.

EVIDENCE-DEFINED BOUNDARY
The Phase 3.11 hand-level evidence was aggregated before intervention. HARD 5–7 against dealer 7–Ace produced 4,667 active DOUBLE hands Primary and 4,737 Holdout, with mean net/hand -1.0015 / -0.9576 and 0.00% immediate bust rate. Against dealer 2–6, the same totals were materially less damaging at -0.3192 / -0.3626. Phase 3.12 therefore tests only the stronger 7–Ace region.

MECHANISM
The hypothesis is not immediate busting. DOUBLE commits 2 units, grants exactly one additional card, then terminates drawing. HIT commits the base unit and retains normal drawing flexibility.

PRIMARY RESULTS
Average ending bankroll — Textbook: 99.956; Selective Aggression: 95.203.
Positive sessions — Textbook: 48.14%; Selective Aggression: 28.26%.
Maximum Selective-Aggression bankroll: 126.0 at Hand 25.

HOLDOUT RESULTS
Average ending bankroll — Textbook: 99.972; Selective Aggression: 95.065.
Positive sessions — Textbook: 47.42%; Selective Aggression: 27.86%.
Maximum Selective-Aggression bankroll: 127.5 at Hand 30.

COMPARISON WITH PHASE 3.11
Average bankroll improves +0.706 Primary and +0.599 Holdout.
Positive-session frequency improves +2.06 pp Primary and +2.16 pp Holdout.
No session finishes at 2x, 3x, 5x or 10x.

PATHWAY INTERROGATION
Under the Phase 3.11 DOUBLE target family, every active target hand ended after exactly one additional player card and had 0% immediate bust rate.
Under Phase 3.12 HIT:
- Primary: 4,632 active target hands; 755 stop after the third card; 3,877 continue to 4+ cards.
- Holdout: 4,804 active target hands; 761 stop after the third card; 4,043 continue to 4+ cards.
Among the three-card stops, 438 Primary and 450 Holdout lose despite remaining non-bust. Thus the mechanism includes low final totals losing to the dealer as well as later-card busts and other outcomes.

INTERPRETATION
Phase 3.12 confirms a second damage mechanism. Maximum Aggression can be harmful even where the extra card cannot immediately bust, because doubling combines larger stake exposure with forced one-card termination and forfeits the flexible continuation available to HIT.

NEXT DESIGN — PHASE 3.13
The remaining evidence is now dominated by exposed HARD 12/13 states against strong dealer cards, while residual low-total states remain outside the 7–Ace boundary. Phase 3.13 should compare aggregate leverage across these two mechanisms and choose only one coherent next family.


======================================================================
19. PHASE 3.13 — HARD 12–13 AGAINST STRONG DEALERS
======================================================================
PURPOSE
Phase 3.13 does not mechanically continue the previous sequence. It first compares the two mechanisms left by 3.12 and then changes one evidence-selected family only.

PRE-INTERVENTION COMPARISON
HARD 12–13 vs dealer 7–Ace under DOUBLE:
Primary 18,075 hands, mean net -0.7296, bust 33.49%.
Holdout 18,245 hands, mean net -0.7604, bust 34.39%.

Residual HARD 5–8 vs dealer 2–6 under DOUBLE:
Primary 4,887 hands, mean net -0.2202.
Holdout 4,737 hands, mean net -0.2681.
Immediate bust 0%.

The HARD 12–13 strong-dealer family therefore had substantially greater exposure and damage.

INTERVENTION
HARD 12 and HARD 13 versus dealer 7–Ace -> HIT instead of automatic DOUBLE. Dealer 2–6 states remain unchanged.

PRIMARY RESULTS
Average ending bankroll — Textbook: 99.956; Selective Aggression: 96.617.
Positive sessions — Textbook: 48.14%; Selective Aggression: 32.98%.
Paired winner — Textbook: 63.42%; Selective Aggression: 34.82%; tie: 1.76%.
Maximum Selective-Aggression bankroll: 127.0 at Hand 25.

HOLDOUT RESULTS
Average ending bankroll — Textbook: 99.972; Selective Aggression: 96.547.
Positive sessions — Textbook: 47.42%; Selective Aggression: 32.08%.
Paired winner — Textbook: 63.88%; Selective Aggression: 33.66%; tie: 2.46%.
Maximum Selective-Aggression bankroll: 127.5 at Hand 30.

COMPARISON WITH 3.12
Average bankroll improves +1.414 Primary and +1.482 Holdout.
Positive-session frequency improves +4.72 pp Primary and +4.22 pp Holdout.
No session finishes at 2x, 3x, 5x or 10x.

INTERPRETATION
The evidence-selected hybrid has now reached approximately 96.6 units average ending bankroll in both cohorts, compared with approximately 100 for Textbook and approximately 79 for the original frozen Maximum Aggression baseline. The programme continues to import only those Textbook decisions justified by replicated finite-shoe evidence.

REMAINING LANDSCAPE
HARD 8 against strong dealers now appears prominently with zero immediate bust risk and Textbook HIT. HARD 13 vs dealer 2 also remains damaging but represents a different mechanism and Textbook STAND action. Soft hands and split-child states remain in the table as well.

NEXT DESIGN — PHASE 3.14
Aggregate and compare the HARD 8 strong-dealer HIT family against remaining weak-dealer mid-total states such as HARD 13 vs 2. Select only one coherent family for intervention.


======================================================================
20. PHASE 3.14 — HARD 8 AGAINST STRONG DEALERS
======================================================================
PRE-INTERVENTION COMPARISON
HARD 8 vs 7–Ace: 2,385 Primary / 2,514 Holdout, mean net -0.6197 / -0.6444, immediate bust 0%.
HARD 13 vs 2–6: 6,218 / 6,147, mean net -0.5294 / -0.5316, bust 38.21% / 38.90%.

INTERVENTION
HARD 8 vs dealer 7–Ace -> HIT. HARD 13 vs 2–6 is deliberately left unchanged for independent testing.

RESULTS
Primary average bankroll: 96.617 -> 96.916 (+0.299).
Holdout: 96.547 -> 96.845 (+0.298).
Primary positive sessions: 32.98% -> 33.56%.
Holdout: 32.08% -> 33.46%.
Peak Selective bankroll: 130.0 at Hand 30 Primary; 127.5 at Hand 30 Holdout.
No 2x/3x/5x/10x finishes.

INTERPRETATION
The smaller but almost identically replicated gain is consistent with a narrower residual damage family and provides early evidence of diminishing marginal returns as the hybrid policy matures.

MULTIPLIER PROGRAMME BOUNDARY
Phase 3.xx continues to optimise decision quality under the frozen 1-unit wager architecture. Peak/multiplier outcomes remain measured but do not drive wager changes. A later programme can freeze the mature policy and investigate bankroll-growth/multiplier objectives without invalidating this policy evidence.

NEXT DESIGN — PHASE 3.15
Re-audit then independently test the preserved HARD 13 vs dealer 2–6 Textbook-STAND family if it remains the strongest coherent candidate.


======================================================================
21. PHASE 3.15 — HARD 13 AGAINST WEAK DEALERS
======================================================================
PURPOSE
Phase 3.15 independently tests the HARD 13 vs dealer 2–6 family deliberately preserved by 3.14.

RE-AUDIT
Primary: 6,210 target hands, mean net -0.5195, bust 38.16%, loss 60.50%.
Holdout: 6,138, mean net -0.5415, bust 39.15%, loss 61.24%.

INTERVENTION
HARD 13 vs dealer 2–6 -> STAND instead of automatic DOUBLE.

RESULTS
Primary average bankroll: 96.916 -> 97.319 (+0.403).
Holdout: 96.845 -> 97.293 (+0.448).
Primary positive sessions: 33.56% -> 35.48%.
Holdout: 33.46% -> 35.34%.
Peak Selective bankroll: 126.5 at Hand 26 Primary; 126.5 at Hand 30 Holdout.
No 2x/3x/5x/10x finishes.

INTERPRETATION
Phase 3.15 independently validates a second late-stage mechanism: against weak dealer cards, HARD 13 automatic DOUBLE adds unnecessary card exposure and produces substantial bust rates. Replacing it with the frozen Textbook STAND action improves both cohorts.

DISTANCE FROM TEXTBOOK
Selective Aggression now averages 97.319 Primary / 97.293 Holdout versus Textbook 99.956 / 99.972. The hybrid remains behaviourally distinct from Textbook outside evidence-selected overrides.

LATE-STAGE LANDSCAPE
The remaining ranking is fragmented across low hard totals, split-child HARD 12/16 states and soft hands. No single broad family now dominates as clearly as earlier phases.

NEXT DESIGN — PHASE 3.16
Perform a family-level aggregate late-stage audit using exposure, aggregate damage, replication, Textbook-action coherence and mechanism. If no remaining coherent family has meaningful programme-level leverage, document diminishing returns rather than forcing another intervention.


======================================================================
22. PHASE 3.16 — LATE-STAGE FAMILY AUDIT / SOFT 13–17
======================================================================
AUDIT
Residual HARD 5–8 vs dealer 2–6: aggregate net -1,112 Primary / -1,180 Holdout.
Split-child HARD 12–16: -738 / -548.
SOFT 13–17 vs dealer 7–Ace: -2,256 / -2,514.

INTERVENTION
SOFT 13–17 vs dealer 7–Ace -> HIT instead of automatic DOUBLE.

RESULT
Primary average bankroll: 97.319 -> 97.632 (+0.313).
Holdout: 97.293 -> 97.683 (+0.390).
Positive sessions: 35.48% -> 36.38% Primary; 35.34% -> 36.28% Holdout.
No 2x/3x/5x/10x finishes.

DIMINISHING-RETURN INTERPRETATION
Smaller or negative future increments do not invalidate the multiplier programme. They define the boundary of useful policy optimisation. A later multiplier programme will inherit the best validated Phase 3.xx policy checkpoint, not necessarily the final experiment number. Rejected late-stage interventions remain evidence but are not incorporated into the frozen multiplier baseline.

NEXT DESIGN — PHASE 3.17
Re-audit residual families and investigate why HARD 16 strong-dealer states remain in the evidence despite earlier HARD 16 protections. Verify first-decision/split-child architecture before any intervention.


======================================================================
23. PHASE 3.17 — CANDIDATE-INSTRUMENTATION CORRECTION / HARD 12
======================================================================
ARCHITECTURAL AUDIT
Phase 3.16's remaining-state report appeared to reintroduce HARD 16 despite its earlier protection. Direct CSV audit found zero HARD 16 strong-dealer hands with Selective first action DOUBLE in both cohorts. Source inspection confirmed the general HARD 16 Textbook mapping remained active.

REPORTING DEFECT
The old remaining-state ranking treated replicated negative realised outcome as remaining policy damage even when Selective and Textbook already chose the same action. Negative outcome alone is not policy divergence.

CORRECTION
The candidate audit now aggregates only records where Selective Aggression first action differs from Textbook first action, before applying exposure and replicated-negative filters.

CORRECTED LEADING FAMILY
HARD 12 vs dealer 2–6:
Primary 6,180 divergent hands, mean net -0.4479, aggregate -2,768, bust 31.49%.
Holdout 6,226, mean net -0.4099, aggregate -2,552, bust 30.24%.

INTERVENTION
HARD 12 vs 2–6 -> corresponding Textbook action: HIT vs 2–3; STAND vs 4–6.

RESULT
Primary average bankroll 97.632 -> 97.992 (+0.360).
Holdout 97.683 -> 97.931 (+0.248).
Positive sessions 36.38% -> 37.60% Primary; 36.28% -> 36.96% Holdout.
No 2x/3x/5x/10x finishes.

INTERPRETATION
The policy architecture was sound; the audit instrumentation needed a stricter definition of unresolved damage. Correcting that definition prevents unavoidable negative outcomes under already-correct actions from being mistaken for policy errors.

NEXT DESIGN — PHASE 3.18
Aggregate only true first-action divergences across residual low hard totals, HARD 9 strong-dealer states and remaining soft-hand families. If meaningful replicated leverage is no longer present, begin formal policy-closure testing.


======================================================================
24. PHASE 3.18 — TRUE-DIVERGENCE AUDIT / HARD 9
======================================================================
AUDIT
Residual HARD 5–8 vs dealer 2–6: 4,898 Primary / 4,765 Holdout; aggregate net -1,100 / -1,044.
HARD 9 vs dealer 9–Ace: 2,594 / 2,532; aggregate net -1,032 / -1,060; mean net -0.3978 / -0.4186.
All remaining SOFT divergences combined were net positive; narrower SOFT 18–19 vs 9–Ace remained negative but with lower aggregate damage.

INTERVENTION
HARD 9 vs dealer 9–Ace -> HIT instead of automatic DOUBLE.

RESULT
Primary average bankroll 97.992 -> 98.167 (+0.175).
Holdout 97.931 -> 98.109 (+0.178).
Positive sessions 37.60% -> 38.36% Primary; 36.96% -> 37.78% Holdout.
No multiplier threshold reached.

INTERPRETATION
The nearly identical +0.175/+0.178 gain validates the intervention but also strengthens the evidence for diminishing marginal returns. The hybrid is now within roughly 1.8 units of Textbook average bankroll.

NEXT DESIGN — PHASE 3.19
Compare the residual HARD 5–8 weak-dealer family against narrower SOFT 18–19 strong-dealer divergence and begin defining a formal closure threshold for permanent policy complexity.


======================================================================
25. PHASE 3.19 — CLOSURE-THRESHOLD AUDIT / RESIDUAL HARD 5–8
======================================================================
TRUE-DIVERGENCE COMPARISON
Residual HARD 5–8 vs dealer 2–6:
Primary 4,879 divergent hands, mean net -0.2156, aggregate -1,052.
Holdout 4,785, mean net -0.2178, aggregate -1,042.

SOFT 18–19 vs dealer 9–Ace:
Primary 1,697 hands, mean net -0.2946, aggregate -500.
Holdout 1,807, mean net -0.2446, aggregate -442.

INTERVENTION
HARD 5–8 vs dealer 2–6 -> HIT instead of automatic DOUBLE.

RESULT
Primary average bankroll 98.167 -> 98.324 (+0.157).
Holdout 98.109 -> 98.378 (+0.269).
Positive sessions 38.36% -> 38.96% Primary.
Positive sessions 37.78% -> 39.30% Holdout.
Peak Selective bankroll 128.0 at Hand 26 Primary / 126.0 at Hand 30 Holdout.
No 2x/3x/5x/10x finishes.

DISTANCE FROM TEXTBOOK
Selective Aggression is now approximately 1.63 units below Textbook Primary and 1.59 units below Textbook Holdout on average ending bankroll.

CLOSURE-THRESHOLD INTERPRETATION
Phase 3.19 remains directionally positive but the gain is small and less tightly matched across cohorts than many earlier phases. From this point a permanent policy rule should require genuine action divergence, adequate exposure, replicated damage, coherent mechanism and enough programme-level improvement to justify added complexity.

Negative or negligible late-stage interventions should be rejected rather than incorporated. The later multiplier programme will inherit the best validated policy checkpoint, not automatically the final experiment number.

NEXT DESIGN — PHASE 3.20
Precisely aggregate the remaining SOFT 18–19 strong-dealer true-divergence family, especially dealer 9/10, and test it only if its expected leverage remains above the emerging closure threshold. Otherwise begin formal policy-freeze validation.


======================================================================
26. PHASE 3.20 — CLOSURE-THRESHOLD TEST / SOFT 18–19
======================================================================
TRUE-DIVERGENCE EVIDENCE
SOFT 18–19 vs dealer 9–Ace:
Primary 1,707 divergent hands, mean net -0.2906, aggregate -496.
Holdout 1,812, mean net -0.2373, aggregate -430.

INTERVENTION
SOFT 18–19 vs dealer 9–Ace -> corresponding Textbook action:
SOFT 18 -> HIT; SOFT 19 -> STAND.

RESULT
Primary average bankroll 98.324 -> 98.412 (+0.088).
Holdout 98.378 -> 98.496 (+0.118).
Positive sessions 38.96% -> 39.32% Primary.
Positive sessions 39.30% -> 40.30% Holdout.
No 2x/3x/5x/10x finishes.

CLOSURE INTERPRETATION
The rule remains genuinely beneficial in both cohorts, but its programme-level effect is now only about one tenth of a bankroll unit. This is strong evidence of diminishing marginal returns.

The hybrid now sits approximately 1.54 units below Textbook Primary and 1.48 units below Textbook Holdout while remaining behaviourally distinct outside evidence-selected overrides.

REMAINING LANDSCAPE
True divergences are now narrow and generally low-leverage, including small soft-hand states, HARD 10 vs Ace/10 and HARD 9 vs dealer 8.

NEXT DESIGN — PHASE 3.21
Begin formal policy-freeze validation with NO new policy intervention. Preserve the full 3.20 checkpoint and test robustness, remaining-divergence leverage, candidate-filter integrity and readiness for multiplier-programme handover. Reopen optimisation only if a materially significant unresolved family is found.


======================================================================
27. PHASE 3.21 — FORMAL POLICY-FREEZE VALIDATION
======================================================================
POLICY CHANGE
None. The complete Phase 3.20 decision architecture was carried forward unchanged.

REPRODUCTION
Phase 3.21 exactly reproduced Phase 3.20:
Primary Selective average 98.412; positive sessions 39.32%.
Holdout Selective average 98.496; positive sessions 40.30%.
Second complete execution was byte-identical.

TRUE-DIVERGENCE AUDIT
All remaining first-action divergences combined were aggregate positive:
Primary 9,454 hands, +1,042 aggregate, +0.1102/hand.
Holdout 9,267 hands, +658 aggregate, +0.0710/hand.

Narrow negative families remain, but their leverage is small or imperfectly replicated. HARD 10 vs 10/Ace was -92 Primary / -46 Holdout across about 2,100 hands each. HARD 9 vs 8 was -78 Primary but +6 Holdout. SOFT 13–16 vs 2–4 was -18 / -110 across about 1,550 hands each.

FREEZE CONCLUSION
No broad, high-leverage replicated divergence comparable with the earlier Phase 3.xx policy defects remains. Further permanent overrides are not justified by current evidence relative to added complexity.

Decision-policy optimisation is therefore CLOSED at the current evidence level.
Official frozen policy checkpoint: Phase 3.20.
Formal freeze-validation experiment: Phase 3.21.

MULTIPLIER HANDOVER
The next research layer must preserve the frozen decision policy. Wager/multiplier experiments may investigate stake architecture but must not silently alter HIT/STAND/DOUBLE/SPLIT policy.

NEXT DESIGN — PHASE 3.22
Establish a multiplier-programme baseline with the frozen Phase 3.20 policy and unchanged 1x wager. Record policy fingerprint, Hand 1–30 bankroll trajectory, peak timing, drawdown/survival and multiplier observations before introducing any wager intervention.


======================================================================
28. PHASE 3.22 — MULTIPLIER PROGRAMME BASELINE (1x ONLY)
======================================================================
PROGRAMME TRANSITION
Decision-policy optimisation remains frozen at Phase 3.20 and validated by Phase 3.21. Phase 3.22 opens the separate multiplier/wager research layer without changing either the policy or wager.

NEW PROSPECTIVE INSTRUMENTATION
Bankroll is recorded after every hand. The experiment now measures mean Hand 1–30 trajectory, per-session peak, peak timing, maximum drawdown, whether 2x/3x/5x/10x bankroll thresholds were reached at any time, and bankroll survival.

BASELINE
Primary average ending bankroll: 98.412.
Holdout: 98.496.
Maximum observed: 126.0 at Hand 29 Primary / 125.5 at Hand 29 Holdout.
Mean session peak: 104.304 / 104.317.
Mean maximum drawdown: 8.317 / 8.256 units.
Ever reached 2x, 3x, 5x or 10x: 0.00% for every threshold in both cohorts.
Ever reached bankroll <=0: 0.00% both cohorts.

TRAJECTORY
Primary mean bankroll: Hand 1 99.901; Hand 10 99.427; Hand 20 98.930; Hand 30 98.412.
Holdout: 99.933; 99.495; 98.975; 98.496.

INTERPRETATION
The 1x baseline demonstrates that multiplier bankroll targets are not merely lost before Hand 30: none is reached during play in these cohorts. Future multiplier work can now quantify reward, timing, drawdown and survival against this baseline.

NEXT DESIGN — PHASE 3.23
Introduce the first wager intervention only: constant 2x base wager, meaning 2 units rather than 1. Preserve the frozen decision policy, 100-unit start and 30-hand horizon. Repeat all Phase 3.22 trajectory, peak, drawdown, survival and multiplier-threshold metrics.


======================================================================
29. PHASE 3.23 — FIRST WAGER INTERVENTION: CONSTANT 2x
======================================================================
CHANGE
Base wager increased from 1 unit to 2 units on every initial hand. The frozen Phase 3.20 decision policy remains unchanged. Normal DOUBLE, SPLIT and blackjack payout mechanics scale proportionally.

RESULT VS PHASE 3.22
Average ending bankroll:
Primary 98.412 -> 96.825 (-1.587).
Holdout 98.496 -> 96.992 (-1.504).

Maximum observed bankroll:
Primary 126.0 -> 152.0.
Holdout 125.5 -> 151.0.

Mean session peak:
Primary 104.304 -> 108.608.
Holdout 104.317 -> 108.633.

Mean maximum drawdown:
Primary 8.317 -> 16.634.
Holdout 8.256 -> 16.512.

No session ever reached 2x, 3x, 5x or 10x bankroll. No session reached bankroll <=0.

POSITIVE-SESSION INVARIANCE
Because Phase 3.23 is pure constant wager scaling with identical decisions/card paths, the sign of every session result is preserved. Positive-session frequency is therefore exactly unchanged from 3.22:
Primary 39.32% at both 1x and 2x.
Holdout 40.30% at both 1x and 2x.

INTERPRETATION
The 2x wager amplifies the existing return distribution rather than improving it. Peak upside grows strongly, but average negative drift and drawdown approximately double. The highest bankroll remains only 152.0 / 151.0, below the 200-unit target.

This establishes that constant wager multiplication is not itself an optimisation strategy.

NEXT DESIGN — PHASE 3.24
Test constant 3x base wagering while preserving every other control. Establish whether the near-linear 1x -> 2x scaling continues before introducing evidence-selected conditional wager timing.


======================================================================
30. PHASE 3.24 — SECOND WAGER INTERVENTION: CONSTANT 3x
======================================================================
CHANGE
Base wager increased to 3 units. Frozen decision policy and deterministic card/decision paths remain unchanged. Direct resolved-hand comparison with 3.23 shows only stake and net fields change.

RESULT
Average ending bankroll:
Primary 95.237; Holdout 95.489.

Positive sessions remain invariant:
Primary 39.32% at 1x, 2x and 3x.
Holdout 40.30% at 1x, 2x and 3x.

Maximum observed:
Primary 178.0; Holdout 176.5.

Mean session peak:
Primary 112.912; Holdout 112.950.

Mean maximum drawdown:
Primary 24.951; Holdout 24.767.

No session reached 200, 300, 500 or 1,000 units at any point. No session reached bankroll <=0.

SCALING CONCLUSION
The 1x -> 2x -> 3x ladder is almost perfectly linear. Constant multiplication scales the same underlying outcomes rather than improving their quality. Upside, negative drift and drawdown all expand proportionally while positive-session frequency remains unchanged.

NEXT DESIGN — PHASE 3.25
Test constant 4x as the first threshold-crossing experiment. The verified Primary maximum is +26 units above start at 1x, so pure scaling projects 100 + 4(26) = 204. Phase 3.25 will determine whether the 200-unit target is actually crossed, how often, when, whether Holdout replicates, whether the gain is retained, and what risk accompanies it.


======================================================================
31. PHASE 3.25 — THRESHOLD-CROSSING EXPERIMENT: CONSTANT 4x
======================================================================
CHANGE
Base wager increased from 3 to 4 units. Frozen policy and deterministic card/decision paths remain unchanged. Direct resolved-hand comparison confirms only stake and net differ from 3.24.

MILESTONE — FIRST 2x BANKROLL CROSSING
Primary: 1/5,000 sessions (0.02%) reached 200; session 3697 reached 204 at Hand 29 and finished Hand 30 at exactly 200.
Holdout: 1/5,000 (0.02%) reached 200; session 3625 reached 202 at Hand 29 but did not retain 200 at the end.

Maximum bankroll: 204.0 Primary / 202.0 Holdout.
No session reached 300, 500 or 1,000.

AVERAGE AND RISK
Average ending bankroll: 93.649 Primary / 93.985 Holdout.
Positive-session frequency remains invariant at 39.32% / 40.30%.
Mean maximum drawdown: 33.268 / 33.023.

For the first time, bankroll exhaustion also appears:
Primary 2/5,000 (0.04%).
Holdout 1/5,000 (0.02%).

INTERPRETATION
Phase 3.25 establishes that 200 units is reachable within the 30-hand horizon. However, attainment is a rare tail event and arrives at the same multiplier level where observed bankroll exhaustion first appears. Constant 4x is therefore boundary evidence, not a successful strategy.

NEXT DESIGN — PHASE 3.26
Test constant 5x once. Determine whether 200 attainment becomes measurably more frequent and quantify the corresponding growth in drawdown and exhaustion. If the same scaling mechanism dominates, use that evidence to decide when to close the simple constant-multiplier ladder and begin evidence-selected wager timing.


======================================================================
32. PHASE 3.26 — CONSTANT 5x: THRESHOLD FREQUENCY & RISK
======================================================================
CHANGE
Base wager increased from 4 to 5 units. Frozen policy and deterministic card/decision paths remain unchanged; only stake and net differ from 3.25.

200-UNIT ATTAINMENT
Primary: 7/5,000 (0.14%) reached 200; 3 finished >=200.
Holdout: 13/5,000 (0.26%) reached 200; 11 finished >=200.
Earliest crossing: Hand 18 Primary / Hand 13 Holdout.

Maximum bankroll: 230.0 / 227.5.
No 300/500/1000 attainment.

RISK
Average ending bankroll: 92.062 / 92.481.
Positive sessions remain invariant: 39.32% / 40.30%.
Mean maximum drawdown: 41.586 / 41.279.

Bankroll <=0:
Primary 40/5,000 (0.80%).
Holdout 26/5,000 (0.52%).
This is a sharp increase from 0.04% / 0.02% at 4x.

INTERPRETATION
The success tail and failure tail expand together. Constant 5x makes 200-unit attainment more frequent but still rare, while exhaustion becomes materially observable. Constant multiplication remains scaling rather than optimisation.

NEXT DESIGN — PHASE 3.27
Perform a boundary analysis before another wager change. Calculate the constant multiplier implied by the verified best route for reaching 300 units and project the associated drawdown/risk scale. Use this evidence to decide whether a direct 300-unit constant-wager boundary test is justified or whether the constant ladder should close in favour of intelligent wager timing.


======================================================================
33. PHASE 3.27 — CONSTANT-LADDER BOUNDARY / BANKROLL-CONSTRAINT AUDIT
======================================================================
CHANGE
None. Constant 5x and the frozen decision policy are retained.

ARCHITECTURAL FINDING
The historical 30-hand simulator records bankroll <=0 but does not stop future wagering. At 5x, 40 Primary and 26 Holdout sessions reached <=0; many subsequently continued and some recovered above zero.

VALIDITY CHECK
No 200-unit success in either cohort occurred after an earlier bankruptcy. Therefore the Phase 3.25/3.26 conclusion that 200 is genuinely reachable remains valid.

BANKROLL-GATED SHADOW
Truncating sessions analytically at first <=0 changes 5x results only slightly:
Primary ending average 92.076; Holdout 92.507.
Positive-session and 200-attainment frequencies remain 39.32% / 40.30% and 0.14% / 0.26%.

BOUNDARY PROJECTIONS
From the verified best 1x route:
300-unit threshold implies approximately 8x.
500 implies approximately 16x.
1,000 implies approximately 35x Primary / 36x Holdout.

INTERPRETATION
These are boundary calculations, not strategy recommendations. Because bankroll exhaustion already appears at 5x, high-multiplier experiments must adopt an explicit bankroll-viability rule before testing 8x or beyond.

NEXT DESIGN — PHASE 3.28
Implement bankroll-constrained wagering and calibrate at 5x. Stop a session when bankroll availability no longer permits continued wagering. Validate the coded result against the Phase 3.27 shadow audit before any 8x / 300-unit boundary experiment.


======================================================================
34. PHASE 3.28 — BANKRUPTCY-STOP ARCHITECTURE CALIBRATION AT 5x
======================================================================
CHANGE
No wager or decision-policy change. Constant 5x retained. Selective execution now stops after the first completed hand producing bankroll <=0; bankroll is fixed at zero and later Selective hands are not dealt.

CALIBRATION
The coded implementation exactly reproduces the Phase 3.27 analytical shadow to reported precision.

Primary: ending average 92.076; positive 39.32%; ever >=200 0.14%; 40 bankruptcy stops; mean max drawdown 41.538.
Holdout: ending average 92.507; positive 40.30%; ever >=200 0.26%; 26 bankruptcy stops; mean max drawdown 41.244.

The stop prevents 164 later Primary and 113 later Holdout hand positions that the historical unconstrained engine would otherwise have simulated.

INTERPRETATION
The post-bankruptcy continuation defect is corrected and validated. Existing 200-unit attainment remains intact because no successful crossing depended on a prior bankruptcy.

LIMITATION
Phase 3.28 is deliberately first-stage. A positive bankroll can still be insufficient to finance a prescribed split or double. This must be measured before high-multiplier testing.

NEXT DESIGN — PHASE 3.29
Retain 5x and add observational affordability instrumentation for base wagers, split commitments and double commitments. Determine whether any current outcomes depend on capital unavailable at the moment of commitment. Do not change policy and do not run 8x until this boundary is understood.

======================================================================
35. PHASE 3.29 — AFFORDABILITY / COMMITMENT AUDIT AT 5x
======================================================================
CHANGE
None. Constant 5x, frozen decision policy and validated bankruptcy-stop architecture retained.

FINDINGS
Primary: 40 unaffordable commitment events across 33/5,000 sessions (0.66%): 10 BASE, 9 SPLIT, 21 DOUBLE.
Holdout: 46 events across 31/5,000 (0.62%): 13 BASE, 9 SPLIT, 24 DOUBLE.

INTERPRETATION
A positive bankroll does not guarantee sufficient capital for the complete prescribed wager commitment. DOUBLE is the largest conflict family in both cohorts. The effect is uncommon at 5x but independently replicated and cannot be ignored before higher multipliers.

RESEARCH SIGNIFICANCE
This separates genuine deliberate risk from impossible financing. Later selective-wager experiments can only be judged credibly if every additional stake is actually fundable.

NEXT DESIGN — PHASE 3.30
Implement explicit affordability enforcement at 5x. Require sufficient capital for base, split and double commitments; define any unaffordable-action fallback explicitly and label it separately. Validate against the 3.29 audit before testing 8x.

======================================================================
36. PHASE 3.30 — AFFORDABILITY ENFORCEMENT CALIBRATION AT 5x
======================================================================
CHANGE
No wager increase. Explicit bankroll-affordability override added.

RULE
A base wager, split or double may occur only when the required capital is available. Unaffordable split leaves the original hand unsplit. Unaffordable double uses textbook non-double fallback; if textbook also requests double, HIT is used. If the base wager itself cannot be funded, the session stops and retains its remaining bankroll.

FINDINGS
Primary: 12 base stops, 7 split fallbacks, 22 double fallbacks.
Holdout: 12 base stops, 5 split fallbacks, 23 double fallbacks.
Independent audit found zero impossible funded commitments in either cohort.

OUTCOMES
Average ending bankroll 92.061 / 92.481.
Positive sessions 39.32% / 40.30%.
Ever >=200 0.14% / 0.26%.
Maximum bankroll 230.0 / 227.5.
The genuine 200-unit evidence remains intact.

INTERPRETATION
The simulator now distinguishes genuine financed risk from impossible financing. This closes the principal bankroll-validity boundary identified in 3.27-3.29.

NEXT DESIGN — PHASE 3.31
Use the validated architecture for the controlled constant 8x / 300-unit boundary experiment. Measure threshold attainment against drawdown, bankruptcy, affordability fallbacks and positive-session probability. Treat 8x as a boundary benchmark, not a recommended strategy, before selective wager timing begins.

======================================================================
37. PHASE 3.31 — CONSTANT 8x / 300-UNIT BOUNDARY EXPERIMENT
======================================================================
CHANGE
Base wager increased from 5 to 8 units under the validated bankruptcy-stop and affordability architecture. Frozen decision policy retained.

BOUNDARY RESULT
300 units is reached in both cohorts:
Primary session 3697 peaks at 308 on Hand 29 and finishes at 300.
Holdout session 3625 peaks at 304 on Hand 29 and finishes at 296.
Frequency is 1/5,000 (0.02%) in each cohort.

200-unit attainment rises sharply:
Primary 5.16% ever / 2.78% ending.
Holdout 5.28% ever / 2.86% ending.

POPULATION COST
Average ending bankroll 87.926 / 88.462.
Positive sessions 39.30% / 40.22%.
Mean maximum drawdown 63.946 / 63.634.
Bankroll reaches zero in 3.14% / 3.20%.

Affordability pressure becomes substantial:
Primary 437 base stops, 159 split fallbacks, 286 double fallbacks.
Holdout 423, 128, 278.

INTERPRETATION
Blind 8x exposure makes large thresholds more visible but widens both tails and pushes many sessions into bankroll constraints. It is a benchmark, not an optimisation strategy.

NEXT DESIGN — PHASE 3.32
Begin high-volume wager-opportunity mapping with no selective wager change initially. Interrogate only information observable before the wager/action: hard/soft class, total, dealer up-card, split-child status, frozen action, hand number, bankroll band and affordability. Require Primary/Holdout replication and rank families by exposure, unit-normalised return, downside, positive-session contribution and association with 200/300 attainment. Use this evidence to choose both WHEN to increase the wager and whether the candidate level should be 2x, 3x, 5x or 8x.

======================================================================
38. PHASE 3.32 — PRE-WAGER OPPORTUNITY MAP / LAGGED EVIDENCE AUDIT
======================================================================
DESIGN CORRECTION
A blackjack wager is committed before the current player cards and dealer up-card are known. Current-hand HARD/SOFT class, dealer up-card, split/action, cards drawn, blackjack and outcome are therefore post-wager diagnostics and cannot select that same hand's wager.

DESIGN
Return to the 1-unit reference wager with no selective multiplier intervention. Map only information available before the next wager: hand number, bankroll, previous outcome, previous player/dealer card counts, previous natural-blackjack flags, previous split and coarse previous-three-hand outcome trend.

FINDINGS
150,000 rounds per cohort were analysed.
Mean unit net per round: -0.052923 Primary / -0.050127 Holdout.
Positive sessions: 39.32% / 40.30%.

With minimum exposure of 200 observations in EACH cohort, zero tested coarse pre-wager families had positive mean next-hand return in both Primary and Holdout.

Previous player/dealer card counts and natural-blackjack status show diagnostic variation but no robust positive replicated next-hand trigger. Coarse winning/losing streaks likewise fail to provide positive replicated evidence.

INTERPRETATION
No multiplier increase is justified yet. Less-negative families are not automatically positive opportunities. This negative result prevents the project from manufacturing a wager rule simply because selective wagering was anticipated.

NEXT DESIGN — PHASE 3.33
Deepen the observational map using pre-declared combinations of completed-history variables while controlling fragmentation and requiring minimum exposure plus independent Holdout replication. Do not introduce 2x until evidence earns it. If robust signal remains absent, retain that conclusion rather than searching indefinitely.

SCOPE NOTE
Shuffle analysis remains an optional final boundary investigation. The project may close after multiplier research if shuffle work would expand scope disproportionately.

======================================================================
END OF GUIDE — CURRENT THROUGH PHASE 3.32
======================================================================
