PHASE 2.02 RESULTS ? CONTINUOUS SIX-DECK BEHAVIOURAL DIVERGENCE VALIDATION RESEARCH QUESTION ----------------- When a textbook controller and a frozen non-textbook aggressive controller begin with the identical finite six-deck shoe, do their different decisions create cumulative shoe-position divergence across a 30-hand session, and does textbook play retain a bankroll advantage? DESIGN ------ Each session constructs one exact deterministically shuffled six-deck shoe. Both profiles start at card 0 of that identical shoe, with 100 units and the same flat 1-unit wager. The shoe is not reset between hands. The textbook controller uses the established basic-strategy decision logic. The non-textbook comparator is the archived Phase 1.29 Maximum Aggression decision policy: split every pair when legal and double every first decision when legal, then use ordinary textbook continuation after the first decision. Neither profile receives remaining-shoe composition or count information. WHY THIS DIFFERS FROM PHASE 2.01 -------------------------------- Phase 2.01 reset the shoe for every independent counterfactual hand. Phase 2.02 allows one decision to change the pointer for the next hand; that new hand can then change later consumption again. The experiment therefore tests propagation of behavioural divergence through a finite inventory. PRIMARY COHORT Sessions per profile: 5000; hands per session: 30 Average ending bankroll ? Textbook / Non-textbook: 99.956 / 78.978 units Average 30-hand net ? Textbook / Non-textbook: -0.044 / -21.022 units Positive sessions ? Textbook / Non-textbook: 2407 (48.14%) / 124 (2.48%) Negative sessions ? Textbook / Non-textbook: 2411 (48.22%) / 4857 (97.14%) Paired session winner ? Textbook higher / Non-textbook higher / tie: 4786 (95.72%) / 194 (3.88%) / 20 (0.40%) Round outcomes T win/loss/push: 65252/71808/12940; N: 45808/92280/11912 Doubles T/N: 15716/154310; splits T/N: 3808/21838 Sessions in which shoe pointers diverged: 5000 (100.00%) Mean first divergence hand, among diverged sessions: 2.674 Hand starts at identical shoe pointer: 23252/150000 = 15.50% Final pointer same / textbook consumed more / non-textbook consumed more: 119 / 234 / 4647 Mean absolute final pointer gap: 12.505 cards Mean cards consumed in 30 hands ? T/N: 165.674 / 177.840 Maximum cards consumed in any session ? T/N: 186 / 206 (shoe size 312) INDEPENDENT HOLDOUT Sessions per profile: 5000; hands per session: 30 Average ending bankroll ? Textbook / Non-textbook: 99.972 / 78.895 units Average 30-hand net ? Textbook / Non-textbook: -0.028 / -21.105 units Positive sessions ? Textbook / Non-textbook: 2371 (47.42%) / 121 (2.42%) Negative sessions ? Textbook / Non-textbook: 2467 (49.34%) / 4867 (97.34%) Paired session winner ? Textbook higher / Non-textbook higher / tie: 4793 (95.86%) / 193 (3.86%) / 14 (0.28%) Round outcomes T win/loss/push: 65178/71701/13121; N: 45810/92474/11716 Doubles T/N: 15412/154170; splits T/N: 3747/21789 Sessions in which shoe pointers diverged: 5000 (100.00%) Mean first divergence hand, among diverged sessions: 2.599 Hand starts at identical shoe pointer: 22786/150000 = 15.19% Final pointer same / textbook consumed more / non-textbook consumed more: 117 / 225 / 4658 Mean absolute final pointer gap: 12.395 cards Mean cards consumed in 30 hands ? T/N: 165.647 / 177.749 Maximum cards consumed in any session ? T/N: 187 / 208 (shoe size 312) INTERPRETATION SAFEGUARDS ------------------------- 1. The non-textbook profile is deliberately fixed before observing Phase 2.02 results. It is not adapted to the new shoe. 2. Flat equal wagering removes progressive betting and walk-away policy as confounders. 3. A session in which one profile finishes ahead does not establish a guaranteed strategy; this is a deterministic paired simulation. 4. Once pointers diverge, later hands are intentionally no longer identical. That is the phenomenon under test, not a failure of control: both profiles began from the identical physical shoe and their own prior decisions caused the divergence. 5. No card-counting inference is permitted. Any finite-shoe effect arises mechanically from depletion and behavioural consumption only. PHASE 2.02 ? FINDINGS SUMMARY Phase 2.02 is the continuous-session companion to Phase 2.01. Both controllers began each session from the same exact 312-card shoe, but the shoe was no longer reset after every hand. PRIMARY: textbook average ending bankroll 99.956 vs non-textbook 78.978; paired mean advantage +20.979 units. Textbook finished above the non-textbook profile in 95.72% of paired sessions; non-textbook finished above textbook in 3.88%. Shoe pointers diverged in 100.00% of sessions, with the first divergence occurring on average at hand 2.674. HOLDOUT: textbook average ending bankroll 99.972 vs non-textbook 78.895; paired mean advantage +21.077 units. Textbook finished above non-textbook in 95.86% and non-textbook above textbook in 3.86%. Pointer divergence occurred in 100.00% of sessions. The principal scientific question is whether the direction and scale of the textbook advantage replicate in holdout while cumulative pointer divergence is widespread. If they do, the finite-shoe extension supports the view that the earlier behavioural finding was not dependent on hand-by-hand reset. If they do not, the discrepancy identifies continuous depletion as an important boundary condition. PHASE 2.02 REPLAY FINGERPRINT: 8568fc513b8dee7e738359cc264d154f7143280aa66af708ae4601cb6a3051a3