DETERMINISTIC BLACKJACK LIVE COMPANION
CLI CONSOLIDATED UAT AND RELEASE HISTORY
Current release: V9.2.52
Updated: 17 September 2026

FINAL PACKAGE NOTE
This is the single retained CLI release/UAT history for the clean V9.2.52 distribution.
Superseded loose V9.2.47-V9.2.52 change-note TXT files have been consolidated here.
The authoritative operational evidence remains in session_evidence and output.txt.

RECENT RELEASE LINE
V9.2.2-UAT  - Preamble-aware S/shuffle analysis work.
V9.2.3-UAT  - Preamble publication handling.
V9.2.4-UAT  - Publication gate correction.
V9.2.5-UAT  - Visible version correction/audit.
V9.2.6-UAT  - Preamble display refinement.
V9.2.7-UAT  - Hybrid counterfactual handling.
V9.2.8-UAT  - Hybrid publication request support.
V9.2.9-UAT  - Session plate structural reference-pack support.
V9.2.10-UAT - Portable publication preflight hard gate. Requires current evidence set and Sessionplate_structure.zip; missing required evidence stops plate creation.
V9.2.11-UAT - Stat-Watching Casual introduced for post-session S analysis using GUI-visible evidence only.
V9.2.12-UAT - Historical Stat-Watching backfill and source-bounded evidence handling; preamble-start basis retained where genuinely linked.
V9.2.13-UAT - Stat-Watching session-story evidence; stay path explicitly remains the same Stat-Watching personality with exit disabled, never Frozen.
V9.2.14-UAT - Session 1 publication exception: no previous Session plate is required for Session 1; no Session 0 may be requested/invented/substituted. Structural precedent comes from Sessionplate_structure.zip while Session 1 evidence remains scientific authority.
V9.2.15-UAT - PERSONAL-mode Stat-Watching post-commit reveal. Human wager/action is committed first. Only after a Personal choice differs from Frozen may the user optionally reveal Stat-Watching Casual's corresponding recommendation and mindset. Stat-Watching remains deterministic, non-learning, and limited to information genuinely visible at that decision point.

V9.2.23-UAT - Structured human-mindset event stream for future replay. Wager deviation reasons are emitted at STAGE WAGER / PRE_DEAL; action-deviation reasons at STAGE ACTION / PRE_OUTCOME; optional final-hand notes are typed as reflection, prospective intention, evidence-quality, counterfactual or mixed and emitted at STAGE POST_HAND / POST_SETTLEMENT_PRE_NEXT_HAND. Same-hand retry retains committed reasoning unchanged. No Frozen/Casual/Stat-Watching/Ultimate-AI policy change.

V9.2.15 UAT / SCIENTIFIC SAFEGUARDS
- Frozen policy unchanged.
- Existing Casual policy unchanged.
- Existing post-session S Stat-Watching analysis unchanged except where explicitly documented by its established release line.
- Personal decision is not contaminated by the Stat-Watching reveal: reveal occurs only after commitment.
- Wager advice cannot use the current dealer upcard, current player cards, hidden dealer card, or future cards because those are unavailable when the wager is committed.
- Play-decision advice may use only information visible at that play-decision point.
- Previously revealed dealer-hole-card observations may contribute only through the frozen Stat-Watching observational rules; the current hidden hole card is never exposed.
- Stat-Watching does not learn from previous successes/failures and does not rewrite its policy. Same visible state produces the same deterministic choice.
- Choosing not to view a recommendation does not change Stat-Watching internal state.
- Genuine linked observed preamble history may seed the visible observational context; it must not be fabricated or appended to an exhausted formal replay.
- Publication evidence gates and Session 1 exception remain in force.

WORKING-FOLDER DISCIPLINE
Keep this consolidated file with the current CLI release. Routine future CLI releases should extend/replace this consolidated history rather than scatter additional historical release-audit TXT files through the CLI root.
Preserve docs_audit as the historical provenance/audit archive.
Preserve session_evidence and output.txt as research evidence/data; they are not installation clutter and are intentionally not bundled into a clean code release package.

PACKAGE CONTENTS
- LiveBlackjackCompanion_PersonalOrFrozen_OptionalHistory_Ledger.java
- Current compiled LiveBlackjackCompanion_PersonalOrFrozen_OptionalHistory_Ledger*.class files
- RUN_CLI.bat
- CLI_V9_2_15_PERSONAL_STAT_WATCHING_REVEAL_AUDIT.txt
- CLI_CONSOLIDATED_UAT_AND_RELEASE_HISTORY_v9_2_15.txt

END OF CONSOLIDATED HISTORY

V9.2.16-UAT
- Added final narrow Stat-Watching five-card 10-value working-memory action extension.
- No change to Frozen, existing Casual, wager trend, exit policy or evidence boundaries.

V9.2.17-UAT
- Tightened the new five-card 10-value memory faculty so it remains a rare attention override.
- Hard 12–16 vs dealer 4/5/6 override: 0/5 recent 10-values only.
- Double 10/11 reluctance override: 4+/5 recent 10-values only.
- Normal Casual-style action remains default; Frozen/Casual/wager/exit rules otherwise unchanged.

V9.2.18-UAT
- Publication-request hardening only; Stat-Watching behavioural baseline from V9.2.17 is unchanged.
- Added mandatory plate legibility requirement for dense tables, decimals, currency, percentages, footnotes and multi-column values.
- Added final visual inspection for ambiguous decimal points, cramped values, clipping, contrast and column alignment.
- Added future-proof supporting-plate rule: universal guides say “Applies to all Session Scorecards” rather than hard-coding a current session range.
- Registered the universal “How to Read the Session Scorecards” and separate “Shuffle Robustness Table — Field Guide” as explanatory publication aids.
- No Frozen, Casual, Stat-Watching wager/action/exit, replay, evidence, bankroll or shuffle-analysis logic changed.


V9.2.18 PACKAGING CLEAN-EXIT AMENDMENT
- RUN_CLI.bat now deletes generated LiveBlackjackCompanion*.class files after the CLI exits.
- Partial/generated class files are also removed if javac fails.
- Loose .class files are no longer shipped in the CLI ZIP; RUN_CLI.bat compiles from the authoritative .java source at launch.
- No Java source, policy, evidence, replay, Stat-Watching, Frozen, or Casual behaviour changed.

========================================================================
SOURCE AUDIT CONSOLIDATED: CLI_V9_2_17_STAT_WATCHING_RARE_OVERRIDE_AUDIT.txt
========================================================================
V9.2.17-UAT — STAT-WATCHING FIVE-CARD MEMORY RARITY TIGHTENING

Purpose
Keep the five-card memory as an occasional human attention override, not a mechanical rejection rule.

Changes from V9.2.16
- Hard 12–16 vs dealer 4/5/6: memory override now requires ZERO 10-value cards among the five most recently remembered visible cards (previously 0 or 1).
- Double on 10/11: reluctance override now requires FOUR OR FIVE 10-value cards among the five most recently remembered visible cards (previously 3+).
- Otherwise normal Casual-style action remains authoritative.
- Existing revealed-hole-card hunch remains unchanged and separate.
- Frozen policy unchanged. Existing Casual policy unchanged. Wager/exit policy unchanged.

Design boundary
Observation is not automatically action. Only a conspicuous five-card extreme can occasionally override the normal hand policy. No new pattern families, prediction, count, shoe model or learning logic were added.

Validation
javac compilation: PASS.


========================================================================
SOURCE AUDIT CONSOLIDATED: CLI_V9_2_18_PUBLICATION_LEGIBILITY_AUDIT.txt
========================================================================
V9.2.18-UAT — PUBLICATION REQUEST / LEGIBILITY AUDIT

PASS: Java source compiles.
PASS: CODE_VERSION = V9.2.18-UAT.
PASS: V9.2.17 Stat-Watching behavioural rules remain unchanged.
PASS: plate-generation request now contains mandatory legibility and final visual-check requirements.
PASS: universal supporting plates are future-proofed and not tied to Sessions 1–8.
PASS: universal scorecard field guide and separate shuffle-table field guide are identified as explanatory aids only.
PASS: no Frozen/Casual/Stat-Watching decision, wager, exit, replay or evidence logic intentionally changed.


========================================================================
SOURCE AUDIT CONSOLIDATED: CLI_V9_2_18_CLEAN_EXIT_CLASS_CLEANUP_AUDIT.txt
========================================================================
CLI V9.2.18-UAT — CLEAN EXIT / CLASS FILE CLEANUP AUDIT
Packaging-only amendment.

RUN_CLI.bat:
- Compiles the authoritative Java source on launch.
- Captures the CLI process exit code.
- Deletes generated LiveBlackjackCompanion_PersonalOrFrozen_OptionalHistory_Ledger*.class files after exit.
- Also deletes partial class output after a compile failure.
- Preserves the CLI exit code.

ZIP packaging:
- Loose .class files intentionally omitted.
- Authoritative .java source retained.

Behavioural change: NONE.
Version remains CLI V9.2.18-UAT.

========================================================================
V9.2.20-UAT — ULTIMATE AI V1 / SESSION 9 BIRTH
========================================================================
Date: 10 September 2026

PURPOSE
- Introduces the final experimental Ultimate AI synthesis layer beginning with publication Session 9.
- Adds AIplayer.txt as the persistent cumulative evidence repository used by the controller.
- Existing Frozen, Casual, Stat-Watching Casual, Personal and Hybrid behaviours remain isolated and unchanged.

CUMULATIVE REPOSITORY
- AIplayer.txt is rebuilt from authoritative completed-session output and stores one provenance-bounded snapshot per publication session.
- Startup prepares the repository through the latest session whose S analysis is COMPLETE.
- After S, the just-completed session is added to the repository for the NEXT session only.
- Temporal gate: Session n can never use its own completed/post-session result to alter earlier decisions in Session n.
- Missing card-level evidence remains N/A rather than being inferred.
- Session 1 therefore retains unavailable exact card/action-derived fields as N/A where the cumulative source cannot support them.

SESSION 1–8 SEED
- The pre-Session-9 repository is generated through publication Session 8.
- Session 2 chronology is taken from the authoritative chronological card block (176 cards) and exposure from committed hand stakes (£645), avoiding stale superseded summary values elsewhere in historical cumulative output.
- Non-publication/correction runs are excluded from the publication-session repository.

ULTIMATE AI V1 — FROZEN WEIGHTED GATE
- Frozen remains the baseline and wins ties, weak evidence and contradictory evidence.
- A different action is possible only when a predecision-safe alternative exists and at least three independent/supporting evidence points converge.
- Wager changes likewise require at least three supporting points.
- Inputs may include current hand state, bounded short-term visible observations, prior revealed-hole memory, cumulative hand-depth/action/wager evidence, bankroll/exposure context and historical exact starting-state outcomes.
- No current hidden dealer card, future card, inferred missing chronology or future/post-session Session-n statistic is available to a decision.
- Scorecard/percentile similarity is retained in the repository/design as long-term context but has ZERO live-decision weight in V1 until a predecision-safe mapping is defined.
- No single curiosity is sufficient to override Frozen by itself.

S INTEGRATION
- For Session 9 onward, S runs the Ultimate AI on the exact supported source basis after existing analysis.
- The AI is evaluated using AIplayer.txt through Session n-1, then Session n is incorporated only after its AI comparison has been written.
- S writes a compact Ultimate AI evidence block with bankroll/completion status, same-basis Frozen comparator, exposure, override counts, path and audit lines.
- Source-exhausted/incomplete AI journeys publish outcome N/A and preserve only the last source-supported bankroll; they are not promoted to completed outcomes.

LIVE / PERSONAL
- Ultimate AI remains silent during live capture in this release.
- No Personal-mode Ultimate AI reveal prompt is added yet; that is deliberately deferred.

VALIDATION
- javac compilation: PASS.
- Startup against the retained cumulative output: PASS.
- Repository horizon: Session 8 / available from Session 9: PASS.
- Publication-session count and temporal gate: PASS.
- Session 2 chronology/exposure correction path: PASS.
- Source-supported dry-run of Ultimate AI engine on retained Session 8/Preamble evidence: PASS (test only; not a Session 9 result).
- Frozen, Casual and Stat-Watching code paths intentionally unchanged.


================================================================
CLI V9.2.20-UAT — DEVIATION PROMPT CLARITY
================================================================
Date: 10 September 2026

- Wager-deviation classification prompt now states:
  [I] intentional deviation (no justification note)
  [O] other (enter justification)
- [O] continues to request a Brief reason/note.
- This is a prompt/annotation clarification only. Wager validation, Frozen policy,
  Stat-Watching, Ultimate AI V1, AIplayer repository logic, settlement, replay and
  evidence boundaries are unchanged.

CLI V9.2.21-UAT — ACTION DEVIATION JUSTIFICATION CLARITY
---------------------------------------------------------
- Hybrid/Frozen-visible action deviations now receive an explicit classification prompt.
- [I] records an intentional action deviation with no justification note.
- [O] records an action deviation and requests a Brief reason/note.
- Applies at every player decision point, including later Hit/Stand decisions within the same hand.
- No Frozen, Casual, Stat-Watching or Ultimate-AI decision rule changed.

================================================================
CLI V9.2.22-UAT — SAME-HAND MINDSET LOCK + FINAL HAND NOTES
================================================================
Date: 10 September 2026

- If CARD STREAM REVIEW is rejected, the factual wager amount remains editable on re-entry.
- A previously captured wager-deviation classification/reason for that same hand is retained verbatim outside the hand rollback and is automatically restored; the operator is not asked to rewrite the mental reasoning after seeing more of the hand.
- Confirmed wager or action deviation while Frozen prompts are visible automatically promotes the observed session classification to HYBRID. Frozen remains the unchanged reference policy.
- After settlement/amendment review, the CLI offers one optional final hand note for retrospective reflection, evidence-quality qualification, or a prospective next-hand intention. The note is timestamped by sequence as captured after settlement and before the next hand.
- No permanent hand-number cue is added to the GUI/live decision environment.
- No Frozen, Casual, Stat-Watching or Ultimate-AI decision rule changed.

VALIDATION
- javac compilation: PASS.
- Same-hand wager reason is held outside HandSnapshot rollback; wager/card facts remain correctable.
- Final-note evidence is written to the authoritative CLI log only after the hand survives amendment review.


================================================================
CLI V9.2.24-UAT — SESSION-END PARSER BOUNDARY FIX
================================================================
Date: 10 September 2026

- Fixes AIplayer repository parsing when Session 9 contains structured human-mindset addendum/event blocks whose headings begin with "END SESSION 9 ...".
- The repository parser now closes a publication session only when the END SESSION marker resolves to the same session ID as the session header (including historical corrected-ID normalization).
- This prevents structured mindset section terminators from prematurely truncating the Session 9 snapshot before its authoritative LEDGER_ENTRY.
- Expected Session 9 repository facts after rebuild: mode HYBRID, hands 30, final bankroll £222.00, P/L £122.00, exposure £730.00, W/L/P 14/14/2.
- No blackjack policy, Frozen, Casual, Stat-Watching, Ultimate-AI gate, mindset-event content, card evidence or shuffle result changed.

VALIDATION
- javac compilation: PASS.
- Session 9 repository rebuild from post-S output: PASS.
- Temporal horizon remains Session 9 / available from Session 10.

========================================================================
CONSOLIDATED CONTINUATION — POST V9.2.24 THROUGH CURRENT V9.2.45-UAT
========================================================================
CONSOLIDATION / PROVENANCE RULE
- This continuation extends the master CLI history rather than keeping routine
  incremental audit TXT files in the active CLI root.
- Entries below are included only where retained source/audit/runtime evidence
  directly supports the stated change.
- Intermediate micro-versions for which no sufficiently specific retained
  release description was located are NOT reconstructed from memory.
- Historical raw evidence remains preserved; later correction/addendum records
  supersede only their explicitly documented analytical/publication scope.

------------------------------------------------------------------------
V9.2.25-UAT — TRANSACTIONAL CARD-COUNT / FUTURE COUNTING REPAIR
------------------------------------------------------------------------
- Retained Session 10 correction evidence identifies a V9.2.24 defect in which
  two rejected H12 dealer-entry cards inflated the raw header card count.
- V9.2.25+ fixes future transactional counting.
- Session 10 historical raw capture remains preserved rather than silently
  rewritten.
- Corrected analytical chronology for Session 10 is separately evidenced as
  170 committed/replayable cards.

BOUNDARY
- This entry records the transactional-counting repair supported by retained
  correction evidence; it does not invent a broader V9.2.25 change list.

------------------------------------------------------------------------
V9.2.26-UAT — NO SEPARATE CHANGE DESCRIPTION CONSOLIDATED
------------------------------------------------------------------------
- A sufficiently specific retained V9.2.26 release/audit description was not
  located in the evidence used for this consolidation.
- No behaviour is inferred for this micro-version.

------------------------------------------------------------------------
V9.2.27-UAT — SESSION 10 H22 APPEND-ONLY EVIDENCE CORRECTION PATH
------------------------------------------------------------------------
- Retained runtime evidence explicitly identifies:
    SESSION 10 H22 EVIDENCE CORRECTION - V9.2.27-UAT
- Raw evidence is preserved.
- Publication/integrity workflow requires the current correction-aware evidence
  path and SESSION INTEGRITY: PASS.
- Rank/suit correction handling remains provenance-bounded rather than silently
  rewriting historical capture.

------------------------------------------------------------------------
V9.2.28-UAT — UNKNOWN-SUIT PUBLICATION / REPLAY RULE HARDENING
------------------------------------------------------------------------
- Retained V9.2.28 runtime/publication text makes the UNKNOWN-SUIT rule explicit:
  a rank-known U/? card remains a legitimate blackjack-replayable card because
  suit does not affect the declared blackjack rules.
- The unresolved physical suit identity remains unresolved and must not be
  guessed or invented.
- Correction-aware SESSION INTEGRITY remains a publication hard gate.

------------------------------------------------------------------------
V9.2.29-UAT — SESSION 10 ULTIMATE-AI TEMPORAL / REPRODUCIBILITY AUDIT
------------------------------------------------------------------------
- Added maintainer audit workflow:
    RUN_SESSION10_ULTIMATE_AI_AUDIT.bat
- Audit requires the corrected Session 10 live report, correction addendum and
  corrected publication shuffle-analysis report.
- It re-runs the Session 10 integrity gate.
- It rebuilds AIplayer.txt to evidence through Session 9 before independently
  regenerating Session 10 Ultimate-AI evidence.
- It proves Session 10 completed outcomes are excluded from the AI decision
  repository used for Session 10.
- It verifies the declared H5 wager override and H15 exit are reproducible.
- It records that the H22 suit correction cannot affect the chosen AI path
  because that AI path terminates at H15.
- AIplayer.txt is restored through Session 10 afterwards, available from
  Session 11.

BOUNDARY
- No blackjack play, wager, strategy, Hybrid, Frozen, Casual, Stat-Watching,
  settlement, shuffle-method, live-input or correction semantics changed.
- The audit does not itself change AI_EXIT/source-completion semantics.

------------------------------------------------------------------------
V9.2.30-UAT — ULTIMATE-AI AI_EXIT COMPLETION SEMANTICS
------------------------------------------------------------------------
- Dedicated Session 10 completion-semantics addendum declares:
    AI_EXIT reached before source exhaustion is a deliberate completed /
    source-supported controller journey.
- SOURCE_EXHAUSTED remains NOT SOURCE-COMPLETE.
- Session 10 Ultimate AI is therefore supported as:
    final bankroll £182.50
    P/L +£82.50
    hands 15
    W/L/P 9/5/1
    exposure £345.00
    termination AI_EXIT
    SOURCE-COMPLETE
- Frozen same-basis comparator remains SOURCE_EXHAUSTED /
  NOT SOURCE-COMPLETE; no final bankroll is inferred.
- H5 wager override and H15 exit remain unchanged.

BOUNDARY
- No wager/action/settlement/cardstream/AI weighting/exit trigger changed.

------------------------------------------------------------------------
V9.2.31-UAT — AUTOMATIC CHATGPT PUBLICATION HANDOFF
------------------------------------------------------------------------
- After successful S analysis the backend automatically creates:
    SESSION_EVIDENCE\ChatGPT_SessionN_Publication_Handoff.zip
- Handoff preserves the evidence hierarchy and may include output.txt,
  AIplayer.txt, dedicated live-session evidence, corrected evidence where
  present, shuffle-analysis evidence, correction/addendum/audit files,
  explicitly linked preamble evidence, Sessionplate_structure.zip,
  required accepted plate reference(s), publication request and handoff
  manifest.
- Missing required publication input is surfaced by the handoff manifest and
  publication must stop rather than substitute/infer missing evidence.
- Automated Session 10 handoff regression/preflight: PASS.
- Embedded V9.2.31 backend compilation: PASS.

BOUNDARY
- No blackjack wager/action/settlement/controller decision policy changed.

------------------------------------------------------------------------
V9.2.32-UAT — S-STAGE PRE-RENDER VERIFIED MANIFEST
------------------------------------------------------------------------
- S writes a session-specific plate-ready verified manifest before the ChatGPT
  handoff ZIP is assembled:
    SESSION_EVIDENCE\SessionN_PreRender_Verified_Manifest.txt
- Candidate publication values are prepared in one controlled S-stage pass,
  including observed/live journey metrics, formal-start Frozen/Casual state,
  synthetic-reference Risk/Journey inputs/outputs where supported,
  Stat-Watching and Ultimate-AI evidence/status, shuffle-robustness values and
  the Session 10+ publication architecture requirements.
- The manifest is a REQUIRED publication-handoff item.
- Session 10 corrected formal-start replay regression retained:
    Frozen: 27 hands | W/L/P 15/10/2 | exposure £480.00 |
            cards 160 | last supported £205.00 | SOURCE_EXHAUSTED
    Casual: 27 hands | W/L/P 15/10/2 | exposure £480.00 |
            cards 160 | last supported £205.00 | SOURCE_EXHAUSTED
    Frozen Risk 24.785 | Journey 52.766
    Casual Risk 24.785 | Journey 52.766
- Standalone javac and isolated Q-shutdown smoke test: PASS.

BOUNDARY
- No blackjack wager/action/settlement/controller policy changed.

------------------------------------------------------------------------
V9.2.33-UAT — RETAINED BUILD-LINE EVIDENCE ONLY
------------------------------------------------------------------------
- A retained GUI build validation states that its production JAR was updated
  with V9.2.33 backend classes.
- The same retained repair record describes the publication-request change as
  an advance to V9.2.34-UAT.
- Because those source statements differ, this consolidation does not invent a
  separate V9.2.33 behavioural change description.

------------------------------------------------------------------------
V9.2.34-UAT — OBSERVED/HYBRID/PERSONAL SCORECARD CONTINUITY
------------------------------------------------------------------------
- S-stage verified manifest independently emits Observed/Hybrid/Personal
  Risk/Journey when the observed path is a complete source-supported 30-hand
  journey.
- Session 7 completed-observed-player scorecard treatment is made an explicit
  structural precedent where applicable while retaining the Session 10+
  five-controller running-P/L bar.
- Publication request explicitly protects actual wager evidence: an observed
  wager outside an intended menu does not invalidate Risk/Journey and must not
  be retrospectively rewritten.
- Session 11 observed-Hybrid regression:
    Risk 86.703 | Journey 69.084
    Survival 74.050 | Path Exposure 81.424 | Recovery Activity 51.777
    maxDD £185 | giveback £135 | exposure £640 |
    absolute movement £580 | 1 recovery episode / 1 successful recovery
- Actual Session 11 H2 £25 wager remains untouched.
- Standalone backend javac and CLI Q smoke test: PASS.

BOUNDARY
- No blackjack wager/action/controller/settlement/card-consumption policy
  changed.
- No Session 11 historical evidence was edited.

------------------------------------------------------------------------
V9.2.35-UAT THROUGH V9.2.44-UAT — MICRO-VERSION DETAIL NOT INVENTED
------------------------------------------------------------------------
- The retained environment demonstrates progression through these backend
  versions, including a V9.2.44 runtime observed after Session 11.
- This consolidated continuation does not assign undocumented behavioural
  changes to individual V9.2.35–V9.2.44 micro-versions.
- Their detailed provenance remains in retained historical/audit/runtime
  material where available.

------------------------------------------------------------------------
V9.2.45-UAT — CURRENT VERIFIED RELEASE / S-STAGE PUBLICATION HARDENING
------------------------------------------------------------------------
- Current verified CLI backend/runtime identity: V9.2.45-UAT.
- Retained release documentation describes the V9.2.45 embedded-backend
  alignment as:
    S-stage Hybrid-detail current-session numeric lock and
    publication-request hard gate only.
- No GUI playability or blackjack decision logic change is attributed to that
  alignment.
- Current cleaned CLI launcher lifecycle:
    compile current source -> run CLI -> remove generated backend .class files
    on exit.
- Current release audit records:
    launch/compile test PASS
    runtime identity PASS
    permanent loose backend classes required: NO
- Ultimate-AI repository current verified state:
    evidence through Session 11
    available from Session 12
- The authoritative CLI folder remains the single evidence/data location.

RELEASE / FOLDER DISCIPLINE
- Keep the active V9.2.45 source at CLI root.
- Keep one authoritative output.txt only in CLI.
- Keep AIplayer.txt and session_evidence in CLI.
- Preserve docs_audit as historical provenance/workflow material.
- Old delivery ZIPs, incremental audits and build residue should not accumulate
  in the active CLI root; quarantine/archive them instead.
- "not needed" is a quarantine/rollback area, not active runtime/evidence.
- This consolidated history is the continuing master CLI release/UAT record.

CURRENT STATUS
- CLI backend: V9.2.45-UAT
- Launch/compile test: PASS
- Runtime identity: PASS
- Evidence repository architecture: retained
- Quick Input reconstructed-card state-reset UAT: GUI-side PASS and therefore
  intentionally NOT recorded as a CLI code correction.


========================================================================
FINAL RELEASE GRADUATION — 14 SEPTEMBER 2026
========================================================================
- UAT COMPLETE / RELEASED / FROZEN THIRD-PARTY BASELINE.
- Runtime identity graduated from V9.2.45-UAT to V9.2.45.
- Release conversion is identity/documentation/build packaging only.
- No blackjack policy, controller, cardstream, settlement, evidence calculation,
  S-stage calculation, replay semantics or historical evidence changed.
- Any future functional change requires a new version.


----------------------------------------------------------------
V9.2.46 — SESSION 11 SHUFFLE-BOUNDARY CORRECTION / PUBLICATION LOCK
----------------------------------------------------------------
- Session 11 raw live capture remains preserved unchanged.
- Operator-confirmed accidental before-H12 shuffle declaration is superseded analytically.
- Corrected Session 11 observed shuffle boundaries: before H1 and before H30 only (2 total).
- Corrected H30 preceding segment: 158 committed cards; total committed cards remains 164; cards after latest genuine shuffle remains 6.
- Corrected analytical Session 11 evidence and a dedicated shuffle-boundary correction addendum are retained in SESSION_EVIDENCE.
- Cumulative output Session 11 analytical/replay block uses the corrected boundary so Historical Replay no longer presents an H12 shuffle.
- S pre-render manifest now emits both explicit observed shuffle count and the complete observed shuffle-boundary list.
- ChatGPT publication request explicitly forbids deriving shuffle counts from previous plate imagery; previous plates remain presentation precedent only.
- ChatGPT handoff automatically includes a session-specific shuffle-boundary correction addendum when present.
- Accepted Session11.png is unchanged because it does not display the erroneous total or H12 boundary.
- No cards, wagers, actions, outcomes, bankroll path, W/L/P, exposure, controller policy, scorecard equations, A-E robustness results, or Replay visual layout changed.

========================================================================
PROJECT-CLOSE ADDENDUM — 15 SEPTEMBER 2026
========================================================================
STATUS
- CLI backend V9.2.46: RELEASED / FROZEN.
- Historical Replay v15.10.195: FINAL OPERATOR UAT PASS / ACCEPTED / FROZEN.
- The project is closed at these release identities.
- RUN_CLI.bat remains the authoritative CLI launcher lifecycle:
    compile current source -> run CLI -> remove generated backend .class files.
- Permanent loose backend .class files are not required in the clean CLI root.

REPLAY / RESEARCH BOUNDARY
- Post-release Replay work through v15.10.195 is presentation/research work and
  does not modify the frozen CLI blackjack/controller mechanism.
- Final Replay UAT accepted normal settlement verdict visibility, split-hand
  vertical containment and child verdict visibility, and pre-deal chip
  presentation.
- SAME STARTING STATE / H-N-L material is descriptive research/presentation
  context only and does not rewrite observed evidence or controller decisions.

ULTIMATE AI V2
- Future/counterfactual research only; NOT implemented in the released engine.
- Ultimate AI V1 remains the frozen implemented experimental synthesis layer.
- Any future V2 H/N/L-context investigation requires a new research/version
  branch and must preserve predecision visibility/temporal-firewall boundaries.

EVIDENCE / CHANGE CONTROL
- Session 11 raw live capture remains preserved.
- V9.2.46 Session 11 analytical shuffle correction remains authoritative:
  observed shuffle boundaries before H1 and H30 only; 2 total; 158 committed
  cards before H30; 164 total committed cards; 6 cards after the latest
  genuine shuffle.
- No project-close documentation update changes cards, wagers, actions,
  outcomes, bankroll path, W/L/P, exposure, controller policy, scorecard
  equations, A-E robustness, S-stage calculations or historical evidence.
- Any later functional or presentation change requires a new version and a
  new UAT cycle.
========================================================================


V9.2.47 — 16 September 2026 — S PUBLICATION HANDOFF REQUEST WORKFLOW ONLY
- S-generated ChatGPT request now explicitly requires the generated request and
  ChatGPT_SessionN_Publication_Handoff.zip to be attached together in one message.
- Added template-reconstruction/not-design hard gate.
- Added explicit benchmark_visual graphical-canvas enhancement execution trigger.
- Added controller/player semantic visual-identity rule aligned with GUI/plates.
- Added Observed Hybrid Risk/Journey visibility/N/A preservation rule.
- Added publication-process learning/tracker gate for future S requests.
- No blackjack policy, wager, action, settlement, bankroll, cardstream, controller,
  replay, shuffle-analysis, score equation or research calculation changed.


17 SEPTEMBER 2026 — GUI PRESENTATION CONTINUITY UPDATE
------------------------------------------------------
Replay v15.10.198 / Live GUI v15.10.139. Main Replay controller identity added; Frozen-vs-Casual compact footer legend aligned with Frozen-vs-Hybrid; human-controller identity added to Live GUI. Presentation only. CLI remains V9.2.47; no evidence or research logic changed.


========================================================================
FINAL CONSOLIDATED RELEASE ADDENDA — V9.2.47 THROUGH V9.2.52
========================================================================

DETERMINISTIC BLACKJACK — CLI V9.2.48
S PUBLICATION HANDOFF REQUEST LOCK
17 September 2026

This revision changes publication-workflow request text only.

The S-generated CHATGPT_SESSION_PUBLICATION_REQUEST.txt is now locked to the
user-UAT gold-standard request. It explicitly requires:
- the generated request and ChatGPT_SessionN_Publication_Handoff.zip together;
- ZIP/session-evidence preflight before rendering;
- verified S evidence/manifest as DATA AUTHORITY;
- accepted plates + benchmark_visual.png as VISUAL/PUBLICATION AUTHORITY only;
- inheritance of the established scorecard family, never freeform redesign;
- specialised Hybrid-vs-Frozen + compact N/A Risk/Journey layout when applicable;
- dense A–E/A–H robustness presentation from verified S evidence;
- benchmark_visual canvas enhancement and final evidence/visual QA.

No blackjack policy, cards, card order, wager/action, settlement, bankroll, Risk/Journey
equations, robustness calculations, controller behaviour or evidence semantics changed.

DETERMINISTIC BLACKJACK — CLI V9.2.47
S PUBLICATION HANDOFF REQUEST UPDATE
16 September 2026

This revision is publication-workflow text generation only.
S now tells the operator to attach BOTH the generated ChatGPT request and the
automatically generated ChatGPT_SessionN_Publication_Handoff.zip in the same message.
It also carries forward the proven Session 12 reconstruction, benchmark_visual canvas
enhancement, semantic controller-icon, Hybrid Risk/Journey and process-tracker gates.

No blackjack play or research calculation changed.

S PUBLICATION RENDERING SAFEGUARD — V9.2.49
17 September 2026

Scope: publication translation only. No captured Session 12 evidence was changed.

Verified existing Session 12 S evidence:
- observed hands: 9
- wager match / above / below: 0 / 5 / 4
- arithmetic lock: 0 + 5 + 4 = 9 PASS

Fix:
- Future S manifests now state the derived observed-wager count explicitly alongside match/above/below.
- The ChatGPT publication request now states that observed wager count is match + above + below and must never be confused with the match count alone.
- Therefore 0 / 5 / 4 renders as "(9 observed wagers)", not "(0 observed wagers)".

No wager, action, bankroll, outcome, card, replay, controller, Risk/Journey, or robustness data logic changed.

S SAME-STATE P/L CORPUS REPAIR — V9.2.50

Scope: post-session S evidence generation only. Live-play capture and blackjack decision/settlement rules are unchanged. Historical Replay is v15.10.208; Live GUI remains v15.10.140.

Defect found:
Session 12 same-state corpus was emitted with the 15-column schema and omitted frozenPnlSum / casualPnlSum, causing Replay AVG P/L / MATCHED HAND to show N/A despite matching paths being present.

Repair:
- HNL collector now accumulates the already-computed per-replay-hand net P/L independently for Frozen and Casual paths.
- Corpus writer restores the established 17-column schema: frozenPnlSum and casualPnlSum are appended.
- Session 12 corpus 20260916_184907 was deterministically regenerated from its retained authoritative captured-card reservoir using the unchanged A-E robustness replay methods and seeds.

Session 12 state 3,3|6 verification:
- matching hands 2643 = Frozen 1293 + Casual 1350
- frozenPnlSum = -40.000000 -> AVG P/L / MATCHED HAND = -£0.03 (rounded to pennies)
- casualPnlSum = -240.000000 -> AVG P/L / MATCHED HAND = -£0.18 (rounded to pennies)

No Replay rendering code changed. No live session evidence was rewritten.

S PUBLICATION HANDOFF — N/A SCORE LAYOUT REFERENCE FIX — V9.2.51
Updated: 17 September 2026

Scope: S-stage ChatGPT publication-handoff packaging only. Blackjack play, settlement, evidence generation, Replay and Live GUI logic are unchanged.

Change:
- Cumulatively increments CLI backend V9.2.50 -> V9.2.51.
- Packages the existing session_evidence/plates/benchmark_na_score_layout.png into generated ChatGPT publication handoff ZIPs.
- The PNG remains a specialised visual/layout reference only for the compact Hybrid-vs-Frozen + Risk Score N/A + Journey Intensity Score N/A component. It is never numeric/session authority.
- When the current observed path is incomplete for Risk/Journey scoring, handoff preflight now requires this reference and reports it missing rather than silently producing an incomplete visual handoff.
- Existing ChatGPT request wording is retained; it already instructs ChatGPT to use the specialised N/A component reference when applicable.

Package hygiene review:
- CLI/session_evidence.zip is not referenced by current Java, BAT, README or operational workflow code.
- Its archived files are duplicated by the live CLI/session_evidence directory; the live directory is newer and contains additional Session 13 evidence and benchmark_na_score_layout.png.
- Therefore session_evidence.zip is omitted from the V9.2.51 distribution package as redundant historical duplication. The live CLI/session_evidence directory remains authoritative and unchanged.

DETERMINISTIC BLACKJACK — S PUBLICATION TRANSFORMATION LOCK — V9.2.52

Scope: publication-handoff instructions and reference packaging only. No blackjack, replay, evidence, scoring, settlement, corpus or GUI logic changed.

Changes:
- Increments CLI cumulatively from V9.2.51 to V9.2.52.
- For incomplete / early-terminated observed Hybrid sessions, S now includes Session12.png as the required primary whole-plate structural precedent when available.
- The generated ChatGPT request now defines an explicit two-stage publication operation:
  1) inherit the correct Session12-like evidence-state structure using current-session S values;
  2) transform that structurally correct plate to benchmark_visual.png publication styling.
- Protects the compact HYBRID vs FROZEN COMPARISON + stacked Risk Score N/A + Journey Intensity Score N/A component as one locked component group during the benchmark styling pass.
- Adds final visual QA checks requiring the Session12-like incomplete-Hybrid structure and protected component group to survive the editorial transformation.
- benchmark_visual.png remains visual/style authority only; it is never numeric/session evidence.
