DETERMINISTIC BLACKJACK DECISION ENGINE RESEARCH
PHASE 2.01 FINDINGS SUMMARY

RESEARCH QUESTION
-----------------
Would the central Phase 1.23 counterfactual result change materially when the non-depleting deterministic rank stream was replaced by an exact deterministic six-deck shoe?

WHAT CHANGED
------------
Only the card-source architecture changed. Each independent hand case now begins from a deterministically shuffled 312-card six-deck shoe with exact finite inventory: 24 Aces, 24 each of ranks 2 through 9, and 96 ten-valued cards. The textbook decision logic and retrospective legal-alternative branching method were preserved. Neither path was given card-counting, remaining-shoe or future-card information.

MAIN FINDING
------------
The result did not change materially.

Phase 1.23 non-depleting rank generator:
- Improvable textbook losing hands: 41.68% primary / 41.76% holdout.
- Alternative branches changing later sequence position: 55.12% primary / 55.00% holdout.

Phase 2.01 deterministic six-deck shoe:
- Improvable textbook losing hands: 41.42% primary / 41.63% holdout.
- Alternative branches changing subsequent shoe position: 55.18% primary / 55.02% holdout.

The six-deck change moved the improvable-loss rate by only -0.26 percentage points in primary and -0.13 percentage points in holdout. The trajectory-divergence rate changed by only +0.06 and +0.02 percentage points respectively.

INTERPRETATION
--------------
This is strong evidence that the earlier 41.7% retrospective finding was not merely an artefact of allowing an unconstrained rank stream. When each hand is instead drawn from a physically valid six-deck inventory, almost the same proportion of textbook losing hands still contains at least one legal alternative that would have produced a better outcome in that exact deterministic case.

This does not mean non-textbook play is superior. The experiment still asks a retrospective question about losing textbook hands, and an alternative is counted as "better" only for that fixed hand and fixed shoe. Phase 1.23 already showed why this must not be converted into a general strategy claim.

WHY THE RESULT IS INTERESTING
-----------------------------
The finite shoe did matter locally: every card removed now reduces a genuine fixed inventory. However, across one million controlled hand cases, that extra realism barely moved the aggregate counterfactual result. The behavioural effect therefore appears substantially more robust to card-source architecture than we had reason to assume before testing it.

DECISION ON NEXT PHASE
----------------------
A single further experiment is scientifically justified, but it should remain brief. Phase 2.02 should move from independent hand-level resets to continuous finite-shoe sessions. Textbook and a frozen non-textbook behavioural profile should begin from the identical deterministic 312-card shoe, with no shoe-awareness or card-counting information. Their own decisions should then be allowed to move them naturally to different positions in the shoe across later hands.

Phase 2.02 would answer a genuinely different question: whether the behavioural comparison remains stable when decision-driven card consumption is allowed to propagate through an entire finite-shoe session.
