PHASE 2.01 RESULTS ? SIX-DECK FINITE-SHOE COUNTERFACTUAL VALIDATION

RESEARCH QUESTION
-----------------
Does the Phase 1.23 retrospective textbook-versus-alternative result remain broadly similar when the non-depleting rank generator is replaced by an exact deterministic six-deck shoe?

DESIGN
------
5,000 primary sessions and 5,000 independent holdout sessions; 100 independent hand cases per session; 1,000,000 original textbook hands in total. Each hand case receives a freshly reset deterministic six-deck shoe containing exactly 312 cards. The textbook hand is played first. Only losing textbook hands are retrospectively branched, and every legal alternative begins from the identical player cards, dealer cards, finite-shoe composition and next-card pointer. No branch receives card-counting or remaining-shoe information.

FINITE-SHOE AUDIT
-----------------
Each shoe contains 24 Aces, 24 each of ranks 2 through 9, and 96 ten-valued cards (10/J/Q/K), totalling 312 cards. Every generated shoe is audited before use.

PRIMARY SIX-DECK COHORT
Hands replayed: 500000
Original textbook losing hands: 240795 (48.16%)
Losing hands with at least one player decision: 217970
Dealer-blackjack or otherwise no-decision losses: 22825
Saved decision points in losing hands: 308696
Legal non-textbook alternative branches tested: 529599
Alternative branches better/equal/worse than textbook original: 153313 / 212087 / 164199
Better alternatives as share of alternatives: 28.95%
Alternative outcomes from original losses ? win/push/loss: 112464 / 12150 / 404985
Decision-robust textbook losing hands: 127694 (58.58%)
Improvable textbook losing hands: 90276 (41.42%)
Converted to at least one win: 72010; push-only improvement: 7275
Player-bust losses improvable: 30161/80510 = 37.46%
Non-bust losses improvable: 60115/160285 = 37.51%
Alternative tests by action ? stand/hit/double: 189209 / 140393 / 199997
Better branches by action ? stand/hit/double: 64630 / 52885 / 35798
Sequence pointer same/earlier/later: 237381 / 174041 / 118177
Alternative branches changing subsequent shoe position: 292218 (55.18%)
Mean absolute shoe-position divergence: 0.686 cards

HOLDOUT SIX-DECK COHORT
Hands replayed: 500000
Original textbook losing hands: 240785 (48.16%)
Losing hands with at least one player decision: 218096
Dealer-blackjack or otherwise no-decision losses: 22689
Saved decision points in losing hands: 307915
Legal non-textbook alternative branches tested: 528909
Alternative branches better/equal/worse than textbook original: 153998 / 210845 / 164066
Better alternatives as share of alternatives: 29.12%
Alternative outcomes from original losses ? win/push/loss: 112687 / 12300 / 403922
Decision-robust textbook losing hands: 127304 (58.37%)
Improvable textbook losing hands: 90792 (41.63%)
Converted to at least one win: 72319; push-only improvement: 7372
Player-bust losses improvable: 30274/80766 = 37.48%
Non-bust losses improvable: 60518/160019 = 37.82%
Alternative tests by action ? stand/hit/double: 188751 / 140228 / 199930
Better branches by action ? stand/hit/double: 64845 / 53289 / 35864
Sequence pointer same/earlier/later: 237924 / 173317 / 117668
Alternative branches changing subsequent shoe position: 290985 (55.02%)
Mean absolute shoe-position divergence: 0.683 cards

COMPARISON TARGET FROM PHASE 1.23
---------------------------------
The archived non-depleting rank-generator experiment reported improvable losing hands of 41.68% in the primary cohort and 41.76% in holdout, with 55.12% and 55.00% of alternative branches respectively changing the future sequence position. Phase 2.01 is a validation comparison against those archived values; it does not overwrite them.

INTERPRETATION SAFEGUARDS
-------------------------
1. A better alternative in one fixed shoe is still a retrospective outcome counterfactual, not proof that the alternative is the superior policy.
2. The textbook controller sees only the player hand and dealer up-card. It receives no shoe composition, count or future-card information.
3. Phase 2.01 resets the shoe between independent hand cases. This deliberately isolates within-hand finite-inventory effects.
4. A later continuous-session experiment would answer a different question because different decisions would move the two profiles to different positions in the same finite shoe.
5. Any decision to continue beyond Phase 2.01 should be based on whether the finite-shoe results materially change or strengthen the archived conclusion.

PHASE 2.01 DETERMINISTIC REPLAY VERIFICATION

Primary seed: 201001
Holdout seed: 201101
Six-deck inventory: 312 cards (24 Aces, 24 each 2-9, 96 ten-valued)
Evidence SHA-256: 075612a4a5176601e0c5f18779fe1cbfa2e528bd13926384df9c3c56e69bd020
A second complete execution must reproduce this exact fingerprint.
