FINAL HISTORICAL-SOURCE AUDIT — V3

HISTORICAL JAVA LINEAGE REVIEWED
- LiveBlackjackCompanion_247RNG
- LiveBlackjackCompanion_247Validated
- LiveBlackjackCompanion_2500PlatformBalance
- Amendments / ResultValidation
- StreamReview
- StreamReview_HoleFirst
- PersonalOrFrozen_HistoryOutcomes
- HistoryOutcomes_SnapshotFix
- HistoryOutcomes_Ledger
- OptionalHistory_Ledger generations supplied by the user

CONFIRMED FROM HISTORICAL SOURCE
- State transitions:
  COOLING -> CAUTIOUS at research-equivalent £120
  CAUTIOUS -> NORMAL at research-equivalent £140
  after reaching £130, bankroll <= £100 enters COOLING.
- Frozen wagers:
  non-NORMAL £15;
  hands 1-8 or bankroll < £130: £15;
  £130-<£150: £20;
  £150-<£170: £30;
  £170+: £15.
- Frozen action table is the same substantive hard/soft/pair table used by V3.
- HARD 11 vs dealer 10 remains the Frozen primary DOUBLE when affordable.
- Pair 10 vs dealer 10 is STAND and is provenance-labelled as the research
  override of the prior aggressive split while also textbook-aligned.
- Platform offset £2400 is protected for live-session affordability.
- Hole-first chronology generation records:
  Player1 -> dealer hole -> Player2 -> dealer up.
- SnapshotFix generation explicitly repaired state/counter restoration.
- Later lineage adds Personal/Frozen mode, history access and cumulative ledger.

TARGETED V3 CORRECTIONS
1. Replay deal-order role assignment corrected to historical HoleFirst evidence.
2. Split affordability is passed into the recommendation layer. When split
   stake is unavailable, the same cards continue through ordinary hard/soft
   policy rather than producing an unavailable SPLIT recommendation.
3. Pending-analysis tracking is per-session and oldest-unresolved-first.
4. lastSavedReport is used only when its filename matches the requested SID.
5. Console separators use ASCII hyphens for Windows-console robustness.

POST-CORRECTION TEST
- exact V3 source compiled with javac -encoding UTF-8.
- Session 3 captured reservoir loaded as 169 cards.
- all A/B/C1/C2/C3/C4/D/E completed 5000 paired replays.
- S appended the analysis and COMPLETE marker.
- runtime in test environment approximately 3.4 seconds, peak RSS about 211 MB.
- standalone report SHA-256:
  3196eac5c39b2d22ceaa664e8897334e3d82a59001c02c238f2f936ad4a933ea

SCIENTIFIC BOUNDARY
No supplied historical Java file contains the old Session 2 Method D shuffle
implementation. V3 Method D is therefore a newly declared, deterministic,
versioned proxy and should be described that way in publication.