BLACKJACK RESEARCH PROGRAMME --- PHASE 3.56+ BRAINSTORM

FINAL EXPERIMENTAL PROGRAMME: CANDIDATE METRICS, QUESTIONS AND RESEARCH
DIRECTIONS

STATUS: PRE-DESIGN BRAINSTORM --- NOT A FROZEN EXPERIMENTAL
SPECIFICATION

PURPOSE This deliberately casts a wide net before Phase 3.56 is
designed. Inclusion here does not mean that a hypothesis is valid or
that every item should become a phase. The final programme should remain
deterministic, reproducible, Primary/Holdout validated, and explicit
about population loss, tail risk and bankroll exhaustion.

1.  RECOVERY AS A FIRST-CLASS RESEARCH OBJECT Measure recovery
    probability after falling below £100; hands to recovery;
    distribution and maximum duration; trough depth; £ recovered; £
    recovered per hand; cumulative wager before recovery; £ recovered
    per £ wagered; repeated below-£100 episodes; repeated recoveries;
    failure to recover; subsequent £150/£170/£200; subsequent £0; and
    Primary/Holdout replication. Key caution: faster recovery may simply
    reflect greater exposure, so duration, cumulative wager, drawdown
    and failure severity must be analysed together.

2.  CUMULATIVE WAGER / EXPOSURE Capture total wager per session;
    mean/median/distribution; exposure by final bankroll band; exposure
    among profitable/losing/break-even/£0/£200 sessions; exposure
    required to reach each £10 tier; exposure between tiers; successful
    versus failed recovery exposure; exposure below £100; exposure in
    conservative/aggressive/protective states; bankroll change per £
    wagered; £200 attainment per unit exposure; bankruptcy per unit
    exposure; and marginal benefit of additional exposure. Central
    question: when does additional wagering stop producing useful target
    reachability and mainly add loss/tail exposure?

3.  EXPOSURE EFFICIENCY Candidate measures: bankroll improvement per
    £100 wagered; £200 probability per £100 wagered; recovery per £100
    wagered; population loss per £100 wagered; bankruptcies per £1,000
    wagered; drawdown per £100 wagered; marginal target probability per
    additional exposure band.

4.  SWEET-SPOT / EXIT RESEARCH Do not assume £200 is automatically the
    ideal exit. Test £110--£200 and finer thresholds around interesting
    regions. For each measure reach probability, retention if
    continuing, hands and cumulative wager to reach, prior drawdown,
    subsequent £0 risk, immediate-walk-away population result, and
    Primary/Holdout stability. Define "sweet spot" in advance as a
    multi-metric balance of attainability, retained gain, exposure,
    drawdown and tail risk.

5.  DRAWDOWN DEEP DIVE Measure drawdown from peak and £100; drawdown
    after tier entry; time below peak; time to regain peak;
    depth-duration pairs; wager during drawdown; repeated drawdowns;
    drawdown before recovery/bankruptcy; drawdown among £200 achievers;
    conditional recovery probability by depth. Investigate the
    apparently interesting £150--£170 region without assuming causation;
    survivor conditioning may matter.

6.  TEXTBOOK ↔ MAX-AGGRESSIVE POLICY ABLATION Keep Phase 3.20 frozen as
    reference. Identify every aggressive action redirected to textbook.
    Restore one decision family at a time to maximum aggression on
    identical deterministic shoes. Measure ending bankroll, positive
    sessions, £200, population return, £0, drawdown, cumulative wager,
    recovery, tails and number of decisions changed. Also consider the
    reverse direction---currently aggressive to textbook---where
    scientifically justified. Only combine changes after isolated
    effects are understood.

7.  INTERACTION TESTING After isolated ablation, test selected
    pairs/small bundles, synergy or antagonism, state-specific effects
    and recovery-state versus profit-deployment policy. Avoid
    combinatorial explosion.

8.  BANKROLL-STATE-SPECIFIC POLICY Possible states: below £100/recovery;
    £100--149 conservative; £150--169 profit deployment; £170--199
    protective; after large win/loss; after recovery; near target. Test
    whether the same textbook/aggressive switch behaves differently by
    state.

9.  WAGER-CONTROLLER OPTIMISATION Study wager slope by tier; £150--169
    deployment strength; aggression/protection transition points;
    protective percentage; gradual versus abrupt transitions;
    caps/floors; recovery-specific caps; post-peak and post-recovery
    wager behaviour.

10. HANDS VERSUS EXPOSURE Distinguish elapsed rounds, wagered rounds,
    cumulative £, cards consumed, split/double branches, and rounds to
    target/recovery/bankruptcy. Determine whether outcomes are
    structured more strongly by time, money exposed or bankroll-state
    path.

11. TARGET REACHABILITY CURVE For every £10 tier record ever reached,
    first-reach hand, first-reach cumulative wager, prior drawdown,
    next-tier progression, fallback, conditional £200 and conditional
    £0. Build a full bankroll transition map.

12. FAILED EXPOSURE Develop and freeze a precise definition before
    measurement. Candidates include wager during failed recovery; wager
    after a peak that is surrendered; wager after tier entry followed by
    falling below tier; or wager producing no durable bankroll
    improvement.

13. PROFIT RETENTION / GIVE-BACK Measure peak bankroll, final bankroll,
    peak-to-finish give-back £/%; give-back after high tiers;
    rounds/exposure between peak and finish; probability of surrendering
    25/50/75/100% of accumulated profit; and alternative-exit effects.

14. SESSION TERMINATION ARCHITECTURE Controlled comparisons: fixed 30
    rounds; immediate £200 stop; alternative targets; drawdown stop;
    peak-profit give-back stop; cumulative-wager ceiling. Test
    individually before combinations.

15. CASINO CHIP DENOMINATIONS --- LATER STAGE Translate the continuous
    controller into an explicitly defined realistic table/chip model.
    Test table minimum, allowed units, rounding rules and inability to
    execute exact percentage wagers. Measure effects on recovery, target
    reach, cumulative wager, drawdown and bankruptcy. Do not assume one
    universal casino denomination set.

16. SHUFFLE ARCHITECTURE --- MAJOR MISSING LINE Separate current
    continuous deterministic six-deck shoe from: cut-card/penetration
    reshuffle; fresh shoe per session; CSM-like replenishment; and
    multiple independently generated deterministic shuffle orders.
    Potential tests: independent shoe replication; starting offsets;
    continuous progression versus session reset; controlled penetration;
    cut-card versus CSM-like architecture; policy robustness across
    shuffle architecture. Never select favourable shoes post hoc.

17. SHOE-POSITION ROBUSTNESS WITHOUT CARD COUNTING Record shoe position
    at important events, stratify early/middle/late shoe, test starting
    offsets and independent shuffles. This is robustness analysis, not
    silently introducing card counting.

18. PHASE 3.35 TIMING CLOSURE Do not casually reopen the closed
    card-history/timing hypothesis. Any later shuffle question must be
    explicitly distinguished from the Phase 3.35 hypothesis.

19. DISTRIBUTIONAL METRICS For major policies report mean, median,
    percentiles, standard deviation, min/max, histograms, tails,
    profitable/losing/break-even proportions and population return.
    Averages alone can conceal the Phase 3.26 frequency-versus-magnitude
    problem.

20. BANKROLL EXHAUSTION / RISK OF RUIN Track £0 probability, bankruptcy
    hand, cumulative wager before bankruptcy, prior peak, drawdown path,
    recovery attempts and bankruptcy conditional on tier/exposure.

21. CONDITIONAL SURVIVOR BIAS Higher tiers contain selected survivors.
    Always distinguish population statistics from "among tier entrants"
    statistics. Do not infer causation merely because high-tier
    survivors look stronger.

22. ROBUSTNESS COHORTS Continue Primary/Holdout and paired deterministic
    comparisons. Important final claims may warrant additional
    independently shuffled deterministic cohorts and predeclared
    replication criteria.

23. COUNTERFACTUAL PAIRED-SESSION ANALYSIS Replay identical shoes under
    Policy A/B; identify first divergence; immediate bankroll
    difference; subsequent wager/action/branching differences; targets
    and bankruptcies created/prevented.

24. DECISION-COST ACCOUNTING For each policy switch record affected
    hands, extra/reduced wager, incremental W/L/P, bankroll effect,
    bankruptcies created/prevented, £200 reaches created/prevented and
    recoveries created/prevented.

25. STREAK/SEQUENCE DESCRIPTORS --- DESCRIPTIVE ONLY Record W/L/P
    sequences and streaks where useful, but do not infer predictive
    power without independent controlled evidence.

26. SPLIT AND DOUBLE EXPOSURE Measure additional £ committed through
    splits/doubles, returns on that exposure, tier-specific
    contribution, recovery contribution and bankruptcy contribution.

27. WAGER CONCENTRATION Maximum original-round stake; maximum total
    stake including branches; share of exposure in largest 1/3/5 wagers;
    share in £150--169; share below £100; concentration among successful
    versus failed sessions.

28. TIME AT RISK Hands below £100; hands below prior peak; hands per
    tier; hands at high stakes; exposure in vulnerable states; time
    £150→£200; time peak→exit/bust.

29. RECOVERY VERSUS FRESH ASCENT Compare high-tier entrants who
    previously fell below £100 with those who did not. Compare
    subsequent £200, drawdown, exposure, bankruptcy and profit
    retention.

30. PATH EFFICIENCY Tier transitions; upward/downward transitions;
    repeated visits; bankroll distance travelled versus net gain; £
    wagered per net £ gained; rounds per net £ gained.

31. NEAR-TARGET £170--£199 Deep-dive £200 probability by tier, fallback,
    exposure to target, give-back, drawdown, current protective stake,
    alternative protection levels and intermediate exits.

32. PROFIT-DEPLOYMENT £150--£169 Determine what the current high-stake
    region actually contributes. Test a predeclared response curve
    around the current deployment percentage rather than assuming the
    current level is optimal.

33. RECOVERY CONTROLLER After recovery anatomy is measured: textbook
    recovery; frozen-policy recovery; reduced-wager recovery; controlled
    aggressive recovery; drawdown-dependent recovery; recovery stop
    rule. Judge by recovery probability AND exposure, duration, survival
    and population return.

34. MULTI-OBJECTIVE EVALUATION Standard comparison panel: ending
    bankroll; population return; positive sessions; £200; £0; max
    drawdown; cumulative wager; recovery efficiency; give-back; exposure
    efficiency. There may be no single "best" policy.

35. PROVISIONAL PROGRAMME ORDER Stage A --- Measurement before
    intervention: recovery duration/efficiency, cumulative wager,
    drawdown/path, failed exposure, give-back. Stage B --- Explain
    current 3.55 architecture: tier transitions, £150--169 deployment,
    near-target protection, recovery paths. Stage C --- Controlled
    policy ablation: textbook→aggressive and aggressive→textbook one
    decision family at a time. Stage D --- Wager/exit optimisation:
    deployment strength, protection, recovery controller, sweet-spot
    exits. Stage E --- Realism/robustness: denominations/table
    constraints; shuffle architecture; penetration/session resets;
    independent-shoe replication. Stage F --- Final validation: fresh
    deterministic cohorts, frozen candidate architecture, full
    risk/exposure report and limitations. Do not allocate all phase
    numbers now. Evidence should determine how many subphases each stage
    deserves.

36. MUST-CAPTURE INSTRUMENTATION FOR 3.56+ Bankroll before/after each
    round; peak/trough; max drawdown; tier entries/exits/transitions;
    original-round stake; total stake including split/double; cumulative
    wager overall/by state; recovery events/duration/exposure; hands
    below £100; target first-reach hand/exposure; give-back; bankruptcy
    hand; selected policy action; whether frozen policy changed
    aggressive action; shoe position/cards remaining; shuffle/reset
    events; termination reason.

37. GUARDRAILS Frequency ≠ magnitude. Target attainment ≠ positive
    expectation. Fast recovery ≠ efficient recovery. High-tier outcomes
    are conditional on survival. Larger wager ≠ better policy. Lower
    survivor drawdown ≠ causal benefit. One favourable deterministic
    shoe ≠ shuffle robustness. Replication ≠ guarantee of future
    gambling performance. No post-hoc selection of shoes, thresholds or
    metrics.

38. ADDITIONAL BACKLOG METRICS Area under bankroll curve; area below
    £100; area below previous peak; largest one-round bankroll shock;
    worst-tail average/expected shortfall; target-to-ruin ratio;
    recovery-to-ruin ratio; conditional £200/£0 by drawdown depth;
    exposure required for final £10/£20/£30 to target; value surrendered
    by continuing beyond candidate exits; path archetypes;
    session-length sensitivity after the 30-round architecture is fully
    characterised.

39. EXTERNAL-RESEARCH OBSERVATIONS TO INVESTIGATE, NOT ASSUME
    Preliminary outside research supports treating shuffle architecture
    as a genuine methodological variable. Conventional cut-card shoes
    and continuous shuffling are distinct; published simulation work
    describes a small cut-card effect. Continuous shuffling can slightly
    reduce per-hand house edge under basic strategy while also
    increasing hands per hour by reducing shuffle downtime. Blackjack
    literature also treats wager size, expected return, variance and
    risk of ruin jointly. These points justify controlled tests; they do
    not predetermine this project's results.

40. PROVISIONAL CENTRAL QUESTION Can the validated Phase 3.20 policy and
    Phase 3.36--3.55 human-like bankroll controller be refined so target
    attainment and successful recovery become more exposure-efficient
    while controlling drawdown, bankruptcy, profit give-back and
    population loss---and do apparent improvements survive policy
    ablation, realistic wager denominations, alternative shuffle
    architectures and fresh deterministic replication?

RECOMMENDED FIRST MOVE Phase 3.56 probably should NOT begin by changing
policy. The strongest opening is likely an instrumentation/measurement
phase exposing recovery duration, cumulative wager, failed exposure,
drawdown path, give-back and tier transitions under the already
validated architecture. This creates a clean baseline for every later
intervention and reduces reruns caused by missing metrics.

41. ACTION-SPACE EXTREME CONTROLS --- STAND-DOMINANT VS HIT-TO-BUST

PURPOSE These are not candidate "good" strategies. They are deliberately
extreme boundary/control policies intended to bracket the player-action
space around the validated Phase 3.20 reference policy.

Conceptual bracket: STAND-DOMINANT EXTREME ← FROZEN PHASE 3.20 POLICY →
HIT-TO-BUST EXTREME

A. STAND-DOMINANT / NO-ADDITIONAL-CARD CONTROL Primary clean boundary: -
after the initial player deal, take no voluntary additional player
cards; - preserve automatic/terminal rule handling such as natural
blackjack; - define splits, doubles and any forced/automatic conditions
explicitly before execution.

Possible secondary variant: - normally take no additional cards; -
permit an additional card only under a precisely predeclared condition
related to the dealer up-card exceeding the player's current total; -
define what happens after that permitted card before execution.

The pure no-additional-card version should be tested first because it
provides the cleanest boundary.

B. HIT-TO-BUST CONTROL - continue taking additional player cards
whenever the experimental rule legally permits; - stop only when the
player busts, reaches a predeclared terminal state such as 21, or
another formal game rule terminates the hand; - natural blackjack,
splits, doubles, 21 and other edge conditions must be explicitly frozen
before coding.

METRICS Run these controls through the full 3.56+ instrumentation: -
average ending bankroll and total population return; - profitable /
losing / break-even sessions; - £200 attainment and bankroll exhaustion
at £0; - player hand-bust frequency; - maximum drawdown and cumulative
wager; - cards consumed and rounds/hands played; - split/double counts
and exposure; - tier occupancy/transitions; - recovery frequency and
successful recovery duration; - failed recovery exposure; -
peak-to-finish give-back; - termination reason; - exact
number/proportion of player decisions differing from Phase 3.20; -
Primary/Holdout replication.

SCIENTIFIC VALUE These controls can help distinguish how much observed
bankroll behaviour is attributable to hand-action policy versus the
wager/bankroll controller layered on top of it.

They also provide strong engine-validation signatures. A stand-dominant
control should produce very different card-consumption and player-bust
behaviour from a hit-to-bust control. If recorded metrics do not reflect
those expected qualitative differences, instrumentation or
implementation should be audited before interpreting bankroll results.

QUESTIONS - Does extreme standing reduce player busts while severely
impairing target progression? - Does hit-to-bust dramatically increase
player busts and tail damage? - How do cumulative exposure and recovery
behave under each extreme? - Where does validated Phase 3.20 sit between
these boundaries? - Do later textbook↔aggressive ablations move outcomes
gradually between the boundaries, or are effects concentrated in a small
number of decision families?

PLACEMENT IN THE PROGRAMME Do not make either extreme policy the initial
Phase 3.56 intervention.

First establish the expanded 3.56 measurement/instrumentation baseline
under the validated architecture. Once those metrics are proven, these
extreme controls become useful benchmark phases for interpreting the
later policy-ablation programme.

GUARDRAIL These are deliberately pathological boundary experiments. A
surprising short-run target percentage must not be interpreted as
evidence that "never hit" or "always hit" is a rational blackjack
strategy. Evaluation remains population-wide and must include return
magnitude, bankroll exhaustion, drawdown, exposure and replication.

42. EVIDENCE-EARNED MULTIPLIER REOPENING / ESCALATION

PURPOSE
Do not assume that the 2x ceiling established by the exploratory human-like
programme must remain permanent, but equally do not reopen larger multipliers
merely because 3x, 4x, 5x or higher wagers are available. The 3.56+ programme
should continuously ask whether newly measured, independently replicated
conditions genuinely EARN the right to test risk above 2x.

CENTRAL QUESTION
Can the richer 3.56+ evidence identify a predeclared circumstance in which
wagering above 2x improves the multi-objective result sufficiently to justify
the additional exposure, rather than simply widening the outcome distribution?

EVIDENCE GATE BEFORE ANY >2x TEST
A multiplier above 2x should only be prospectively tested after a candidate
condition has emerged from the later research and has survived the relevant
replication standard. Candidate evidence may arise from recovery efficiency,
drawdown state, cumulative wager/exposure, bankroll tier, profit deployment,
near-target protection, sweet-spot analysis, denomination constraints,
shuffle/penetration robustness, policy ablation, or another predeclared 3.56+
metric. Discovery alone is not permission to escalate.

SEQUENTIAL ESCALATION
If a replicated condition appears genuinely favourable at 2x, test the next
higher multiplier prospectively and in isolation. A possible ladder is
2x -> 2.5x -> 3x, with any further increase considered only after the preceding
step earns continuation. Do not jump directly back to the earlier blind
3x/4x/5x/8x constant-wager experiments.

REQUIRED COMPARISON
Every candidate escalation should be compared against the same condition at
1x and 2x, preferably by paired deterministic replay. Measure at minimum:
average/median ending bankroll; population return; profitable/losing/break-even
sessions; target attainment; bankroll exhaustion; maximum drawdown; cumulative
wager; exposure efficiency; recovery efficiency; failed exposure; profit
give-back; tail outcomes; affordability/denomination effects; and
Primary/Holdout or fresh-cohort replication.

ROLLBACK PRINCIPLE
The higher multiplier must be rejected if the apparent benefit is explained
mainly by greater exposure, survivor conditioning, larger rare wins, or a
widened distribution while population return, drawdown, bankruptcy, failed
exposure or other risk measures deteriorate beyond the predeclared acceptance
boundary. Failure at a higher multiplier returns the architecture to the last
supported level; it does not invite post-hoc retuning until a favourable result
appears.

RELATIONSHIP TO PHASES 3.22--3.35
This does NOT reopen the closed card-history/timing hypothesis. Phases
3.22--3.35 established that blind constant scaling and the tested
evidence-derived previous-hand history signals did not justify aggressive
multiplier escalation. The new question is different: can the much richer
bankroll-path, exposure, recovery, policy, denomination and shuffle evidence
generated in 3.56+ identify a robust STATE or CONTROL CONDITION under which
additional risk is justified?

POSSIBLE FINAL OUTCOMES
Both outcomes are scientifically valuable:
1. Evidence eventually supports a narrowly defined >2x condition, which then
   survives prospective replication and becomes part of the final candidate
   architecture.
2. No condition earns escalation, providing strong end-of-programme evidence
   that 2x is the practical multiplier ceiling within the tested architecture.

GUARDRAIL
The principle remains unchanged: RISK IS EARNED BY REPLICATED EVIDENCE, NOT BY
HOPE. A higher target-reach percentage alone is insufficient. Any >2x
escalation must survive the full population-wide risk/exposure evaluation.

PROGRAMME PLACEMENT
Do not make this an early Phase 3.56 intervention. First establish the expanded
measurement baseline and allow the recovery, drawdown, cumulative-exposure,
policy-ablation, wager-controller, denomination and shuffle investigations to
generate the evidence. Multiplier reopening is a later conditional branch,
activated only if those experiments provide a defensible reason to test it.

43. CAPITAL-ALERT THRESHOLD SENSITIVITY --- LATER VALIDATION BRANCH

PURPOSE
The current CP10 / bankroll <£50 Capital Alert is a validated state within the tested
architecture, but it must not be treated as the only possible definition of severe
distress. Later research may prospectively test alternative thresholds such as <£70
when a defensible objective-state reason exists.

SCIENTIFIC RULE
Do NOT sweep £50, £55, £60, £65, £70, etc. and select whichever produces the most
attractive result. Any alternative threshold must be predeclared before execution,
compared with the frozen <£50 reference on identical deterministic populations, and
then replicated on fresh cohorts.

REQUIRED MEASURES
For each candidate threshold record:
- incidence and first-trigger hand;
- cumulative wager/exposure at trigger;
- subsequent recovery to £100 / £150 / £170 / £200;
- subsequent £0;
- continuation versus walk-away paired outcomes;
- exposure removed/added;
- maximum drawdown and tail outcomes;
- objective-state trade-off for Capital Preservation, Balanced Recovery and
  Opportunity Preservation.

POSSIBLE ARCHITECTURAL OUTCOME
Research may support multiple severity levels rather than one universal threshold,
for example an earlier WARNING state and a later SEVERE CAPITAL ALERT. This must be
earned by prospective evidence, not imposed by terminology.

44. PHASE 3.130 VISUAL / DECISION-FRAMEWORK REUSE

The Phase 3.130 three-layer architecture should be retained as a reusable comparison
framework where appropriate:
OBSERVED CURRENT STATE × CURRENT OBJECTIVE × HISTORICAL EVIDENCE.
Future metric branches can reuse the same visual grammar while changing the tested
state/intervention. Useful quantitative plates may add population n, transition
probability, typical hand and cumulative wager/exposure at state entry. Do not add
these fields mechanically when they reduce clarity.

The anti-future-information boundary remains mandatory: completed-session labels,
future cards and future bankroll outcomes cannot be leaked into live decisions.

45. PHASE 3.132 PAIRED £150 ACTION DECOMPOSITION — COMPLETED

Phase 3.132 froze the three Phase 3.131 actions and replayed identical deterministic
sessions pairwise. No 25% threshold optimisation was performed.

Four-cohort result:
- HARD_EXIT_150 affected 47.81% of sessions; mean final delta +£2.685; mean exposure
  removed £55.315; mean max drawdown avoided £10.165; 288 £0 outcomes prevented;
  4,028 £200 outcomes removed and 40 preserved.
- TRAIL_150_GIVEBACK_25 affected 23.28% of sessions; mean final delta +£1.598; mean
  exposure removed £31.243; mean max drawdown avoided £6.711; the same 288 £0
  outcomes were prevented; 1,115 £200 outcomes were removed and 2,953 preserved.

Interpretation: the £150 objective-state trade-off is now supported by paired-session
evidence, not aggregate means alone. Hard exit buys maximum protection at very large
opportunity cost; the frozen trailing action remains a reproducible middle position.

NEXT: Phase 3.133 should characterise the anatomy of help-versus-harm divergence before
any threshold refinement. Preserve the anti-future-information boundary: divergence
labels may be used retrospectively for research, never as live decision inputs.

46. PHASE 3.133 HELP-VERSUS-HARM DIVERGENCE ANATOMY — COMPLETED

The paired divergence established in 3.132 was characterised at the actual
intervention point using observable hand, bankroll, peak, cumulative exposure
and drawdown variables. The 25% trailing threshold remained frozen.

Hard exit HELPED/HARMED groups differed modestly at intervention but diverged
strongly afterwards. Trailing HELPED/HARMED groups likewise showed descriptive
differences in timing, bankroll, peak, exposure and drawdown. These are retrospective
associations only.

CRITICAL GUARDRAIL: HELPED/HARMED is a completed-replay research label and must
never become a live input. Phase 3.133 does not establish future predictability.

NEXT: Phase 3.134 should be a prospective discrimination gate. Any candidate
observable rule must be predeclared from the 3.133 anatomy and tested on fresh
deterministic cohorts, with no same-population optimisation and no 25% threshold tuning.

47. PHASE 3.134 PROSPECTIVE EXPOSURE DISCRIMINATION GATE — COMPLETED

A single observable discriminator was predeclared before fresh-cohort execution:
cumulative wager at the frozen 25% trailing trigger <=£225.

The gate was tested on four new deterministic 5,000-session cohorts
(204501 / 204601 / 204701 / 204801). No threshold search or policy change occurred.

FOUR-COHORT RESULT
<=£225:
- n=3,772
- HELPED 52.76%, HARMED 20.33%, NEUTRAL 26.91%
- HELPED:HARMED ratio 2.595
- mean trailing final delta +£4.917

>£225:
- n=2,789
- HELPED 36.29%, HARMED 26.21%, NEUTRAL 37.50%
- HELPED:HARMED ratio 1.384
- mean trailing final delta +£2.602

The predeclared directional hypothesis was supported independently in all four
fresh cohorts. This establishes prospective discrimination, not a complete action
rule. The high-exposure band still contains more HELPED than HARMED sessions.

NEXT: Phase 3.135 should test an exposure-gated trailing ACTION POLICY on another
fresh population: permit the frozen 25% trailing stop only when trigger exposure
is <=£225; otherwise continue. Compare against frozen continuation and always-trail.
Do not tune £225 or 25%.


48. PHASE 3.135 EXPOSURE-GATED TRAILING ACTION TEST — COMPLETED

Fresh cohorts tested continuation, always-trailing, and trailing only when trigger
exposure was <=£225. The gate and 25% threshold were frozen.

Result:
- Always trailing retained the strongest protection metrics.
- Continuation retained the strongest £200 opportunity.
- The exposure-gated action preserved more £200 opportunity than always trailing
  but surrendered most of its protection benefit and tracked continuation closely.

Conclusion: the <=£225 variable is a validated discriminator but not a sufficient
stand-alone action policy. Do not immediately tune the threshold.

NEXT: Phase 3.136 should investigate state completeness — whether exposure plus one
additional observable variable materially improves discrimination before any new
action-rule or threshold search.


49. PHASE 3.136 MULTIVARIATE STATE-COMPLETENESS SCREEN — COMPLETED

No policy was changed. The frozen Phase 3.134 fresh-cohort trigger population was
screened to determine which single additional observable variable most improves
the <=£225 exposure discriminator.

Four-cohort HELPED:HARMED ratios:
- exposure alone: 2.595
- + trigger hand <=8: 2.959
- + bankroll >=£140: 3.564
- + peak >=£165: 2.902
- + drawdown <=£25: 2.941

Current bankroll is the strongest candidate second state variable. This is discovery,
not validation, because the same population selected the candidate.

NEXT: Phase 3.137 must freeze exposure <=£225 AND bankroll >=£140 and test the
two-variable discriminator unchanged on entirely fresh cohorts before another
action-policy experiment.


50. PHASE 3.137 FRESH TWO-VARIABLE DISCRIMINATOR VALIDATION — COMPLETED

The Phase 3.136 candidate was frozen:
exposure <=£225 AND bankroll >=£140 at the frozen trailing trigger.

Four new 5,000-session deterministic cohorts were used. No cutpoint tuning occurred.

Exposure-only fresh total:
n=3,844; HELPED:HARMED 2.458.

Two-variable fresh total:
n=1,978; HELPED 77.20%; HARMED 22.80%; HELPED:HARMED 3.386.

The two-variable ratio exceeded exposure-only independently in all four cohorts:
M 3.298, N 3.310, O 3.200, P 3.823.

The discovery ratio was 3.564, so fresh validation is slightly weaker but clearly
replicated.

IMPORTANT NEW QUESTION:
The two-variable subgroup has stronger HELPED frequency enrichment but lower mean
final delta (+£3.409) than exposure-only (+£4.366). Frequency and magnitude are
therefore not interchangeable objectives.

NEXT: Phase 3.138 should characterise HELPED versus HARMED monetary magnitude and
tails inside the validated state before another action-policy experiment.


51. PHASE 3.138 FREQUENCY × MAGNITUDE ASYMMETRY — COMPLETED

The prospectively validated two-variable state from 3.137 was audited for monetary
magnitude rather than frequency alone.

State retained unchanged:
exposure <=£225 AND bankroll >=£140 at the frozen trailing trigger.

Across 1,978 selected fresh-cohort sessions:
- HELPED 1,527 (77.20%), mean +£22.485, median +£15;
- HARMED 451 (22.80%), mean -£61.181, median -£60.
Mean HARM magnitude is 2.721× mean HELP magnitude.

Aggregate contribution:
HELP +£34,335.00
HARM -£27,592.50
NET +£6,742.50 (+£3.409 per selected session).

The magnitude asymmetry is therefore real, but it does not erase aggregate paired
benefit. Net contribution remained positive in all four fresh validation cohorts.

NEXT: Phase 3.139 is now scientifically justified as a whole-population fresh action
test of the two-variable gated trailing rule versus continuation and always-trailing.
Freeze £225, £140 and 25%; do not retune.


52. PHASE 3.139 TWO-VARIABLE GATED TRAILING ACTION TEST — COMPLETED

The validated two-variable discriminator was finally embedded as a live sequential
action policy on four entirely fresh 5,000-session cohorts.

Frozen rule:
at the 25% trailing trigger, stop only if exposure <=£225 AND bankroll >=£140.

Whole-population result:
CONTINUE: avg final £85.239; £200 20.56%; £0 38.49%; maxDD £81.237; exposure £343.592.
ALWAYS TRAIL: avg final £86.426; £200 14.69%; £0 37.06%; maxDD £74.729; exposure £312.835.
TWO-VARIABLE GATED: avg final £83.946; £200 21.79%; £0 40.11%; maxDD £86.069; exposure £369.166.

Paired gated vs continuation:
HELPED 2,700; HARMED 1,921; NEUTRAL 15,379.
Aggregate HELP +£104,460; HARM -£130,315; NET -£25,855.

Therefore the discriminator does NOT transfer directly into a superior action policy.
The gated policy is rejected in its current form.

CRITICAL RESEARCH LESSON:
A validated counterfactual state discriminator is not automatically a valid sequential
control policy. Applying/declining interventions changes the later state distribution.

Do not tune £225 or £140.

NEXT: Phase 3.140 policy-induced state drift audit. Instrument accepted, declined and
later/re-entry trailing triggers and locate where the adverse magnitude is generated
before designing another intervention.


53. PHASE 3.140 POLICY-INDUCED STATE DRIFT AUDIT — COMPLETED

6,537 sessions encountered a frozen trailing trigger.
1,950 accepted at first trigger; 4,587 declined.

Accepted-first branch remained favourable versus reference:
paired net +£4,110.

Declined-first branch generated the 3.139 reversal:
paired net -£29,965.
After decline, sessions accumulated another 11.617 hands and £245.615 exposure on average.
Only 4.64% later reached a qualifying trigger despite 44,484 later trigger events.

OBJECTIVE-AWARE FINDING:
The declined gated pathway is not simply "worse".
It reaches £200 in 30.96% versus 15.61% for reference, but has much greater capital risk:
30.30% finish below £70 and 13.32% at £0.

Therefore average bankroll must remain a mandatory comparator but must NOT become the
universal objective function. Future profile conclusions should explicitly distinguish:
capital preservation; target seeking; positive-session outcome; exposure/drawdown;
and aggregate economic efficiency.

NEXT: 3.141 post-decline pathway stratification, with no new action rule.


54. PHASE 3.141 POST-DECLINE PATHWAY STRATIFICATION — COMPLETED

The 4,587 declined-first sessions were separated into mutually exclusive pathways:

- TARGET_PROGRESS: 1,420 (30.96%)
- CAPITAL_PRESERVED: 1,225 (26.71%)
- MID_LOSS: 552 (12.03%)
- SEVERE_LOSS: 779 (16.98%)
- TERMINAL_ZERO: 611 (13.32%)

Combined target/preservation pathways = 57.67%.
Combined loss pathways = 42.33%.

The path anatomy becomes increasingly adverse with additional hands, additional
exposure and time below £100. Target progress averaged 7.608 extra hands, £187.965
extra exposure and 0.813 hands below £100; severe loss averaged 16.561 extra hands,
£331.330 exposure and 11.100 hands below £100.

This reinforces objective-specific interpretation:
continuation after decline creates genuine target opportunity, but a sizeable
lower-tail branch experiences prolonged destructive exposure.

NEXT: Phase 3.142 early post-decline checkpoint analysis (+3/+5 hands), no policy change.
Test whether observable post-decline state evolution separates target/preservation
from loss pathways before any new action design.


55. PHASE 3.142 EARLY POST-DECLINE CHECKPOINT SIGNAL DISCOVERY — DESIGNED

Predeclared checkpoints: +3 and +5 hands after first declined trigger.
GOOD retrospective label = TARGET_PROGRESS + CAPITAL_PRESERVED (57.67%).
LOSS retrospective label = MID_LOSS + SEVERE_LOSS + TERMINAL_ZERO (42.33%).

Checkpoint observables: bankroll, incremental exposure, hands below £100, repeated
triggers, regain of first-trigger bankroll, post-decline peak and trough.

Data-integrity guard: Phase 3.141 did not retain exact +3/+5 snapshots, so these
must be explicitly generated by instrumented replay and cannot be inferred from
final pathway averages.

No policy or threshold changes. Any candidate signal must be frozen and
fresh-validated before action testing.


56. PHASE 3.143 INSTRUMENTED +3/+5 CHECKPOINT REPLAY — COMPLETED

The exact checkpoint data deliberately left unfilled in 3.142 were generated by
instrumented deterministic replay.

Coverage:
+3 = 4,089 / 4,587 declined sessions (89.14%).
+5 = 3,620 / 4,587 (78.92%).

Strongest continuous separator at both checkpoints: CURRENT BANKROLL.
+3 GOOD £137.919 vs LOSS £103.344; d=0.954.
+5 GOOD £133.056 vs LOSS £91.363; d=1.067.

Secondary separation:
post-decline trough d=0.784 (+3), 0.872 (+5);
post-decline peak d=0.759, 0.773;
hands below £100 |d|=0.633, 0.737;
later triggers |d|=0.670, 0.680.

Regained first-trigger bankroll:
+3 GOOD 57.17% vs LOSS 25.72%.
+5 GOOD 52.29% vs LOSS 18.07%.

Incremental exposure itself separates weakly (d=0.128 / 0.155), suggesting the
state consequence of exposure matters more than exposure quantity alone.

IMPORTANT:
+5 separation is stronger but checkpoint coverage is lower, so survivorship/availability
must remain explicit.

NEXT: Phase 3.144 fresh-validation of the checkpoint RELATIONSHIP, not optimisation
of a new bankroll threshold. Predeclare higher current bankroll, greater regain
frequency and higher trough in GOOD than LOSS at +3/+5.


57. PHASE 3.144 FRESH CHECKPOINT RELATIONSHIP VALIDATION — COMPLETED

Four fresh 5,000-session cohorts U/V/W/X tested the three relationships frozen
from 3.143 at both +3 and +5 hands.

ALL 6 PREDECLARED DIRECTIONAL TESTS PASSED.

+3:
bankroll GOOD £136.782 vs LOSS £104.361 (d=0.915);
regain 57.50% vs 25.65%;
trough £112.485 vs £94.523 (d=0.759).

+5:
bankroll GOOD £133.536 vs LOSS £91.876 (d=1.076);
regain 53.62% vs 17.96%;
trough £102.257 vs £80.231 (d=0.841).

Current-bankroll effect size reproduced especially closely:
3.143 discovery d=0.954/+3 and 1.067/+5;
3.144 fresh validation d=0.915/+3 and 1.076/+5.

Coverage also reproduced:
fresh +3 88.02%, +5 77.78%;
discovery +3 89.14%, +5 78.92%.

Incremental exposure again separates weakly, reinforcing that resulting bankroll/path
state is more informative at these checkpoints than exposure quantity alone.

IMPORTANT:
Do not convert group means into stop thresholds.

NEXT: 3.145 distribution/overlap mapping of current bankroll and regain status,
including checkpoint availability, before any candidate intervention boundary is frozen.

FUTURE RESEARCH NOTE RETAINED:
Alternative hit/stand/double/split hand-decision policies may later be studied by
paired deterministic replay, but the current behavioural/state evidence chain should
be completed and frozen first.


58. PHASE 3.145 CHECKPOINT DISTRIBUTION AND OVERLAP MAPPING — COMPLETED

Using the fresh Phase 3.144 checkpoint population, current bankroll was mapped into fixed £20 descriptive bands.

GOOD outcome probability rises monotonically with bankroll at both +3 and +5 checkpoints.
Examples:
+3: <£60 15.54%; £100-119.99 46.15%; £140-159.99 68.11%; £180-199.99 89.96%.
+5: <£60 11.38%; £100-119.99 46.92%; £140-159.99 68.60%; £180-199.99 86.89%.

Regain status also separates strongly:
+3 regained 71.42% GOOD vs not-regained 38.93%.
+5 regained 73.02% vs 33.89%.

Central 90% bankroll overlap remains broad:
+3 approximately £77.50–£157.50.
+5 approximately £70.00–£152.50.

Therefore no single bankroll mean/threshold should be promoted directly into action.
The next candidate-state work should combine bankroll band + regain status and limit the number of candidate states to avoid threshold mining.

NEXT: 3.146 candidate combined-state discovery with coverage/risk/objective guardrails; still no live intervention.


59. PHASE 3.146 CANDIDATE COMBINED-STATE DISCOVERY — COMPLETED

Threshold-mining guard: only three broad bankroll regions (<£100, £100–139.99, >=£140)
were crossed with previously validated regain status.

HIGH + REGAINED:
+3 n=1,105; GOOD 80.90%; target £200 56.92%.
+5 n=814; GOOD 81.70%; target £200 60.44%.
This is the principal opportunity-favouring candidate state.

LOW + NOT REGAINED:
+3 n=1,119; GOOD 28.78%; target 10.19%; severe 27.61%; zero 27.17%.
+5 n=1,452; GOOD 24.86%; target 6.54%; severe 29.06%; zero 28.93%.
This is the principal capital-risk candidate state.

MIDDLE remains mixed:
not-regained GOOD ~48%; regained GOOD ~57–59%.
Do not force it into a binary decision.

Rare discordant states (HIGH/not-regained and LOW/regained) are retained descriptively
but are too small to promote as principal candidates.

NEXT: 3.147 fresh validation of the three-state architecture. No intervention yet.

FUTURE OBJECTIVE RETAINED:
Alternative hit/stand/double/split policies remain a later robustness extension after
the current state/intervention evidence chain is frozen.


60. PHASE 3.147 FRESH THREE-STATE ARCHITECTURE VALIDATION — COMPLETED

Fresh cohorts Y/Z/AA/AB, 20,000 sessions total.
Frozen state definitions from 3.146.
ALL 6 PREDECLARED TESTS PASSED.

+3:
OPPORTUNITY GOOD 80.65%; target 60.86%; severe+terminal 13.22%.
MIDDLE GOOD 52.63%; target 26.51%; severe+terminal 32.96%.
CAPITAL_RISK GOOD 29.11%; target 8.56%; severe+terminal 55.96%.

+5:
OPPORTUNITY GOOD 82.28%; target 62.19%; severe+terminal 11.29%.
MIDDLE GOOD 52.80%; target 23.41%; severe+terminal 31.21%.
CAPITAL_RISK GOOD 24.74%; target 7.13%; severe+terminal 61.36%.

This is the clearest three-state separation yet:
OPPORTUNITY = target potential dominates.
MIDDLE = genuinely unresolved.
CAPITAL_RISK = deterioration dominates.

Important causal guard:
classification does NOT prove stopping/continuing is beneficial.

NEXT: 3.148 state-entry counterfactual value mapping before executing any intervention.
Quantify what capital could be frozen at state entry versus what continued play later produces,
and quantify opportunity surrendered by premature stopping in OPPORTUNITY states.


61. PHASE 3.148 STATE-ENTRY COUNTERFACTUAL VALUE MAP — COMPLETED

FREEZE NOW was evaluated as a counterfactual accounting benchmark against the unchanged continued path.

Mean continue-minus-freeze delta was negative in all three states.

+3:
OPPORTUNITY freeze £170.985 vs continue £166.432; delta -£4.553; £200 60.86%; £0 3.42%.
MIDDLE freeze £118.862 vs continue £110.832; delta -£8.030; £200 26.51%; £0 12.78%.
CAPITAL_RISK freeze £77.104 vs continue £67.431; delta -£9.673; £200 8.56%; £0 28.61%.

+5:
OPPORTUNITY delta -£4.072; £200 62.19%; £0 3.05%.
MIDDLE delta -£7.448; £200 23.41%; £0 10.30%.
CAPITAL_RISK delta -£8.863; £200 7.13%; £0 30.82%.

KEY INTERPRETATION:
OPPORTUNITY continuation sacrifices some average retained bankroll but creates very high target attainment.
CAPITAL_RISK continuation sacrifices average capital while producing little target attainment and high terminal risk.
MIDDLE remains mixed.

This confirms the user's earlier observation that average bankroll must not become the universal outcome function.

NEXT: 3.149 formalise separate CAPITAL PRESERVATION, TARGET-SEEKING and BALANCED objective views before any live intervention.


62. PHASE 3.149 OBJECTIVE-FUNCTION FORMALISATION — COMPLETED

Three objective views are now explicit:

CAPITAL PRESERVATION:
focus on retained checkpoint bankroll, terminal-zero risk and exposure.
Accounting preference favours FREEZE in all tested states because mean continue-minus-freeze is negative.

TARGET SEEKING:
focus on £200 attainment and continuation upside.
Opportunity quality is strongly state-dependent:
OPPORTUNITY ~61–62% £200;
MIDDLE ~23–27%;
CAPITAL_RISK ~7–9%.

BALANCED:
no opaque composite score.
Report mean give-back, £200, £0, extra exposure and probability continuation exceeds freeze side-by-side.

At +5:
OPPORTUNITY: -£4.07 mean give-back; £200 62.19%; £0 3.05%; extra exposure £135.03.
MIDDLE: -£7.45; £200 23.41%; £0 10.30%; exposure £218.57.
CAPITAL_RISK: -£8.86; £200 7.13%; £0 30.82%; exposure £189.27.

This formalises STATE × OBJECTIVE × COUNTERFACTUAL VALUE and prevents average bankroll
from becoming the universal success function.

NEXT: 3.150 objective-specific intervention specification. Draft/freeze exact observable
inputs, checkpoint timing, permitted action and success metrics for capital-preservation,
target-seeking and balanced policy intents before fresh execution.


63. PHASE 3.150 OBJECTIVE-SPECIFIC INTERVENTION SPECIFICATION — COMPLETED

No execution occurred.

Frozen states:
OPPORTUNITY = bankroll >=£140 AND regained first-trigger bankroll.
CAPITAL_RISK = bankroll <£100 AND NOT regained.
MIDDLE = all other states.

Primary checkpoint: +5 hands.
Secondary checkpoint: +3 hands.

Frozen candidate intents:
CAPITAL PRESERVATION:
FREEZE in CAPITAL_RISK.

TARGET SEEKING:
CONTINUE unchanged in OPPORTUNITY.

BALANCED:
CONTINUE in OPPORTUNITY;
FREEZE in CAPITAL_RISK;
MIDDLE remains unresolved/observe.

No wager escalation or card-play changes are permitted.
Future success metrics must be reported separately by objective.

NEXT: 3.151 fresh intervention execution of CONTINUE_REFERENCE,
ALWAYS_TRAIL_150_GIVEBACK_25, CAPITAL_PRESERVATION_STATE_POLICY and
BALANCED_THREE_STATE_POLICY on entirely new cohorts.


64. PHASE 3.151 FRESH FROZEN INTERVENTION EXECUTION — COMPLETED

20,000 fresh sessions AC/AD/AE/AF.

Primary +5 Capital Risk freeze versus reference:
mean final +£0.628;
terminal £0 -2.275 pp;
exposure -£13.240;
max drawdown -£2.439;
£200 -0.515 pp;
positive sessions -1.600 pp.
State freeze activated in 1,411 / 20,000 = 7.06%.

Secondary +3:
mean final +£0.605;
£0 -1.730 pp;
exposure -£12.212;
maxDD -£2.350;
£200 -0.535 pp.
State freeze 5.65%.

+5 therefore currently gives the cleaner capital-preservation trade-off.

Always-trail comparator is substantially more intervention-heavy:
mean final +£2.570 vs reference, £0 -3.110 pp, maxDD -£11.503,
but £200 -7.210 pp.

Balanced and Capital-Preservation arms are identical BY DESIGN:
Opportunity=continue and Middle=observe do not change reference execution; only Capital Risk=freeze does.

NEXT 3.152:
fresh replication of frozen +5 Capital Risk intervention plus paired changed-session mechanism analysis.
Do not tune thresholds.


65. PHASE 3.152 FRESH REPLICATION + CHANGED-SESSION MECHANISM — COMPLETED

Second fresh 20,000-session population AG/AH/AI/AJ.

Frozen +5 Capital Risk policy replicated extremely closely:
mean final +£0.606 vs reference;
£0 -2.240 pp;
£200 -0.520 pp;
exposure -£13.664;
maxDD -£2.501.
State freezes 7.09%.

Changed-session mechanism:
1,334 sessions changed.
HELPED 845; HARMED 489.
Aggregate HELP +£41,919.38.
Aggregate HARM -£29,797.50.
Net +£12,121.88.
Changed-session median delta +£25.

Terminal £0 prevented / created: 448 / 0.
Severe <£70 prevented / created: 434 / 135.
£200 lost / created: 105 / 0.

This establishes a coherent objective trade-off:
capital preservation gains are real and replicated, but target opportunity is not free.

NEXT: 3.153 changed-session entry-state anatomy. Compare prevented-zero, saved-capital,
harmed/recovery-forgone and lost-£200 cases using observable checkpoint variables.
No intervention refinement yet.


66. PHASE 3.153 CHANGED-SESSION ENTRY-STATE ANATOMY — COMPLETED

Discovery-only anatomy of the replicated +5 CAPITAL_RISK intervention using the existing 3.152 paired cohorts.
No policy change.

Observable checkpoint separation:
PREVENTED_ZERO n=448: mean bankroll £61.00; median £62.50; 42.19% below £60; 12.28% at £90-99.
LOST_200 n=105: mean bankroll £80.81; median £85.00; 11.43% below £60; 40.00% at £90-99.

Prevented-zero cases also show deeper mean max drawdown (£104.79) and more negative recent-five balance (-2.98).
Lost-£200 cases occur earlier on average (hand 14.23).

Direction is clear but distributions overlap materially. No new bankroll threshold is justified.

NEXT 3.154:
multivariable separation audit using only checkpoint-observable variables. Discovery only; no intervention refinement and no future leakage into live decisions.


67. PHASE 3.154 MULTIVARIABLE SEPARATION AUDIT — COMPLETED

Discovery-only audit of PREVENTED_ZERO (448) versus LOST_200 (105) from the Phase 3.153 anatomy evidence.
All 63 non-empty subsets of six checkpoint-observable variables were evaluated with leave-one-cohort-out validation across AG/AH/AI/AJ.

Bankroll alone: AUC 0.759843, fold range 0.751148–0.795767.
Best subset, bankroll + exposure: AUC 0.760215, only +0.000372 over bankroll alone.
Bankroll + recent-five balance: AUC 0.760045, only +0.000202.
Bankroll alone ranked third of all 63 subsets.
Full six-variable model: AUC 0.738021, worse than bankroll alone.

Conclusion: secondary anatomy variables do not materially improve stable separation. Do not create a complex live rule. This is useful negative evidence against premature complexity.

NEXT 3.155:
simple separator robustness audit using a small pre-specified family of interpretable bankroll-led bands/rules. Report prevented-zero capture and lost-£200 capture by cohort. Discovery only; no policy change.


68. PHASE 3.155 SIMPLE SEPARATOR ROBUSTNESS AUDIT — COMPLETED

Phase 3.154 rejected premature multivariable complexity.
3.155 therefore audited four simple checkpoint-bankroll candidate regions across FRESH_AG/AH/AI/AJ.

PREVENTED_ZERO vs LOST_200 capture:

<=£50: 35.49% vs 6.67%; separation +28.82 pp.
<=£60: 46.88% vs 12.38%; separation +34.49 pp.
<=£70: 61.38% vs 20.95%; separation +40.43 pp.
<=£80: 78.79% vs 46.67%; separation +32.13 pp.

Cohort robustness:
<=£60 separation is exceptionally stable: +33.65 to +35.35 pp.
<=£70 gives the strongest overall separation and remains positive in every cohort: +35.35 to +46.58 pp.
<=£80 captures too much LOST_200 opportunity and is less stable.

No threshold is promoted into intervention.

NEXT 3.156:
Freeze <=£60 as conservative/high-specificity candidate and <=£70 as broader capital-risk capture candidate.
Fresh validate both on new cohorts before any live refinement.


69. PHASE 3.156 FROZEN SIMPLE-SEPARATOR FRESH VALIDATION — COMPLETED

Fresh AK/AL/AM/AN, 20,000 sessions.
Frozen candidates from 3.155 evaluated observationally at +5.

Outcome mechanism population:
PREVENTED_ZERO 438.
LOST_200 113.

<=£60 fresh:
PZ 45.66%; LOST_200 10.62%; separation +35.04 pp.
Discovery: 46.88 / 12.38 / +34.49.

<=£70 fresh:
PZ 62.56%; LOST_200 23.01%; separation +39.55 pp.
Discovery: 61.38 / 20.95 / +40.43.

Both candidates retained positive separation in every fresh cohort.
The objective ordering replicated:
<=£60 more selective/target-protective.
<=£70 broader terminal-risk capture.

Important qualification:
the extremely narrow cohort stability of <=£60 in discovery did not itself replicate.
The overall signal did.

No live refinement yet.

NEXT 3.157:
freeze prospective intervention variants for original Capital Risk, <=£60 refinement and <=£70 refinement before execution.


70. PHASE 3.157 PROSPECTIVE SIMPLE-REFINEMENT INTERVENTION SPECIFICATION — COMPLETED

No execution.

Fresh-validated descriptors from 3.156 are now converted into frozen prospective policy specifications.

Four arms for 3.158:
1. CONTINUE_REFERENCE.
2. ORIGINAL_CAPITAL_RISK_FREEZE:
   +5, bankroll <£100 and first-trigger bankroll not regained.
3. REFINED_60:
   original Capital Risk plus checkpoint bankroll <=£60.
4. REFINED_70:
   original Capital Risk plus checkpoint bankroll <=£70.

£60/£70 are nested refinements; they do not redefine CAPITAL_RISK.

No wager/card-play/trigger/gate changes.

Predeclared whole-population metrics:
mean/median bankroll, £200, £0, positive sessions, final <£70, exposure, maxDD, hands, activation.

Predeclared paired metrics:
better/worse/equal, aggregate help/harm/net, £0 prevented/created, £200 lost/created,
severe <£70 prevented/created, exposure and maxDD changes.

Objective interpretation remains separate:
capital preservation / target seeking / balanced.

NEXT 3.158:
fresh four-arm execution on identical new deterministic cohorts.
No threshold tuning.


71. PHASE 3.158 FRESH FOUR-ARM REFINEMENT EXECUTION — COMPLETED

Fresh AO/AP/AQ/AR, 20,000 sessions, +5, contract frozen in 3.157.

Reference:
mean £83.660; £200 21.38%; £0 40.03%.

Original Capital Risk:
mean +£0.668 vs reference;
£0 prevented 471;
£200 lost 102;
activation 7.52%;
exposure -£13.883;
maxDD -£2.580.

Refined <=£60:
mean +£0.166;
£0 prevented 225;
£200 lost only 8;
activation 2.08%;
exposure -£2.864;
maxDD -£0.556.

Refined <=£70:
mean +£0.287;
£0 prevented 300;
£200 lost 21;
activation 3.25%;
exposure -£4.978;
maxDD -£0.959.

The predeclared objective ordering appeared prospectively:
ORIGINAL = strongest capital protection / greatest target cost.
<=£60 = highest specificity / lowest target cost.
<=£70 = intermediate protection and target cost.

Important:
final<£70 is not equivalent to terminal-risk protection because refined policies can freeze and preserve a nonzero bankroll while still ending inside the <£70 region.

NEXT 3.159:
fresh four-arm replication and cohort-level mechanism robustness.
No tuning.


72. PHASE 3.159 FRESH FOUR-ARM ORDERING REPLICATION — COMPLETED

Second fresh live-intervention population AS/AT/AU/AV, 20,000 sessions.

Original:
£0 prevented 420; £200 lost 117; activation 7.43%; net paired +£7,763.75.

<=£60:
£0 prevented 209; £200 lost 10; activation 2.22%; net paired +£3,173.75.

<=£70:
£0 prevented 268; £200 lost 25; activation 3.29%; net paired +£3,806.25.

The 3.158 objective ordering replicated:
Original = maximum tested capital protection / maximum target cost.
<=£60 = lowest target cost / selective protection.
<=£70 = intermediate.

Critically, the same ordering held independently in AS, AT, AU and AV.
All intervention/cohort net paired bankroll contributions were positive.
No terminal-zero or £200 outcomes were created relative to reference.

NEXT 3.160:
pooled replication / precision audit across 3.158 + 3.159 while preserving population identity.
No new policy or threshold tuning.


73. PHASE 3.160 POOLED REPLICATION / PRECISION AUDIT — COMPLETED

No new sessions. Consolidated prospective 3.158 + 3.159:
40,000 sessions across eight fresh cohorts.

Pooled Original:
activation 7.47%;
£0 prevented 891;
£200 lost 219;
net paired +£21,130;
exposure -£14.106/session;
maxDD -£2.519/session.

Pooled <=£60:
activation 2.15%;
£0 prevented 434;
£200 lost 18;
net paired +£6,492.50;
exposure -£3.025/session;
maxDD -£0.585/session.
50.52% of activations prevent terminal £0; 2.10% lose £200.

Pooled <=£70:
activation 3.27%;
£0 prevented 568;
£200 lost 46;
net paired +£9,537.50;
exposure -£5.198/session;
maxDD -£0.971/session.

Ordering remains:
Original = broadest/max capital protection/max target cost.
<=£60 = most selective/min target cost.
<=£70 = intermediate.

Activation rates and mechanism ordering are highly reproducible between 3.158 and 3.159.
Mean-bankroll benefit is positive but more variable in magnitude.

NEXT 3.161:
cross-population boundary stability audit around frozen £60 and £70.
No threshold tuning or new policy.


74. PHASE 3.161 CROSS-POPULATION BOUNDARY STABILITY AUDIT — COMPLETED

Instrumentation finding:
3.158/3.159 prospective exports did not preserve checkpoint bankroll.
No reconstruction or unsupported direct boundary claim was made.

Boundary anatomy therefore uses 3.156 fresh-validation evidence, where checkpoint_bank is explicitly recorded.

Capital-Risk band anatomy:
50-55: PZ 23.81%, LOST200 4.76%.
55-60: PZ 32.35%, LOST200 7.35%.
60-65: PZ 37.82%, LOST200 5.88%.
65-70: PZ 28.71%, LOST200 6.93%.
70-75: PZ 20.54%, LOST200 8.04%.
75-80: PZ 25.82%, LOST200 7.69%.

Exact £60: 14 PZ / 4 LOST200.
Exact £70: 15 PZ / 5 LOST200.

Conclusion:
no evidence of an isolated exact-value artefact.
£60/£70 are interpretable cut-points through a broader bankroll-risk gradient.
No threshold change.

NEXT 3.162:
fresh four-arm prospective execution with checkpoint-bankroll instrumentation added before intervention.
Close the 3.158/3.159 instrumentation gap while retaining frozen thresholds/policies.


75. PHASE 3.162 PROSPECTIVE BOUNDARY-INSTRUMENTATION REPLICATION — COMPLETED

Fresh AW/AX/AY/AZ, 20,000 sessions.
Frozen four-arm +5 contract unchanged.
Checkpoint bankroll now explicitly exported before intervention.

Policy ordering replicated for a third fresh live population:
Original: 441 £0 prevented / 103 £200 lost / activation 7.32%.
<=£60: 207 / 17 / 2.03%.
<=£70: 267 / 33 / 3.07%.

Prospective Capital-Risk boundary bands:
50-55: PZ 35.29%, LOST200 5.88%.
55-60: 36.76%, 8.82%.
60-65: 30.65%, 9.68%.
65-70: 25.88%, 4.71%.
70-75: 29.09%, 5.45%.
75-80: 25.70%, 7.26%.

Exactly £60: 17 PZ / 6 LOST200 among n=49.
Exactly £70: 12 / 1 among n=39.

Conclusion:
boundary signal is regional, not an isolated exact-value artefact.
Local bands are not perfectly monotonic, so do not portray £60/£70 as mathematically unique optima.
The prospective instrumentation gap from 3.161 is closed.

NEXT 3.163:
cohort-level and broader-region robustness audit using the newly instrumented prospective evidence.
No threshold changes.


76. PHASE 3.163 INSTRUMENTED COHORT / REGIONAL ROBUSTNESS — COMPLETED

No new sessions.

Using fresh prospective 3.162 instrumentation, broader Capital-Risk regions:

<=£60:
PZ 51.11%; LOST200 4.20%.

£60-70:
PZ 28.71%; LOST200 7.66%.

£70-80:
PZ 26.99%; LOST200 6.57%.

£80-100:
PZ 17.11%; LOST200 9.09%.

<=£60 remains strongly enriched in each cohort:
AW 54.74/3.16;
AX 43.81/4.76;
AY 53.27/1.87;
AZ 53.06/7.14
(PZ% / LOST200%).

Policy ordering also held in every AW/AX/AY/AZ cohort:
Original > <=£70 > <=£60 for activation, £0 prevention and target cost.

Conclusion:
regional gradient interpretation survives broader aggregation and cohort decomposition.
No threshold change.

NEXT 3.164:
one final fully independent fresh holdout with frozen policy and instrumentation.
Primary holdout: policy objective ordering.
Secondary: <=£60 enrichment versus £80-100.
If reproduced, close/consolidate this branch and return to brainstorm.


77. PHASE 3.164 FINAL INDEPENDENT HOLDOUT / BRANCH CLOSURE — COMPLETED

Fresh BA/BB/BC/BD, 20,000 sessions.
Frozen four-arm +5 architecture.
Checkpoint instrumentation retained.
No tuning.

PREDECLARED HOLDOUT: 4/4 PASS.

Primary policy ordering:
Original > <=£70 > <=£60
for activation, terminal-zero prevention and £200 target cost.

Holdout counts:
Original: activation 7.35%; PZ 427; LOST200 113; net paired +£11,450.
<=£60: activation 2.10%; PZ 194; LOST200 11; net +£2,426.25.
<=£70: activation 2.99%; PZ 247; LOST200 20; net +£3,225.

Secondary regional holdout:
<=£60 PZ 46.30%, LOST200 2.63%.
£80-100 PZ 17.08%, LOST200 12.81%.
PASS.

Conclusion:
the £60/£70 Capital-Risk refinement branch is sufficiently proven for consolidation.
Original = maximum tested capital protection / highest target cost.
<=£60 = selective / lowest target cost.
<=£70 = intermediate.
Thresholds remain cut-points through a gradient, not unique optima.

NEXT:
3.165 recommended as branch consolidation / return-to-brainstorm phase rather than another threshold experiment.
