PHASE 3.16 — LAYMAN RESULTS EXPLANATION
LATE-STAGE FAMILY AUDIT AND SOFT-HAND INTERVENTION

WHY 3.16 STARTED WITH AN AUDIT
By 3.15 the remaining damage was fragmented. Rather than selecting the most negative individual row, Phase 3.16 first aggregated coherent families across both Primary and Holdout.

THE THREE AUDITED FAMILIES
1. Residual HARD 5–8 vs dealer 2–6
Primary: 4,866 hands, mean net -0.2285, aggregate net -1,112.
Holdout: 4,748 hands, mean net -0.2485, aggregate net -1,180.
Immediate bust: 0%.

2. Split-child HARD 12–16
Primary: 1,439 hands, mean net -0.5129, aggregate net -738.
Holdout: 1,404 hands, mean net -0.3903, aggregate net -548.

3. SOFT 13–17 vs dealer 7–Ace
Primary: 5,819 hands, mean net -0.3877, aggregate net -2,256.
Holdout: 5,776 hands, mean net -0.4352, aggregate net -2,514.
Immediate bust: 0%.

The soft-hand strong-dealer family therefore had the largest replicated aggregate damage among these coherent late-stage candidates.

WHAT CHANGED
Phase 3.16 changes ONE family only:
SOFT 13–17 versus dealer 7–Ace now HIT instead of Maximum Aggression automatically DOUBLING.

The corresponding Textbook action across this tested family is HIT.

RESULT
Primary average ending bankroll rises from 97.319 in 3.15 to 97.632 in 3.16: +0.313 units.
Holdout rises from 97.293 to 97.683: +0.390 units.

Positive sessions rise from 35.48% to 36.38% Primary and from 35.34% to 36.28% Holdout.

The intervention therefore improves both independent cohorts.

DISTANCE FROM TEXTBOOK
Textbook:
- Primary 99.956.
- Holdout 99.972.

Selective Aggression:
- Primary 97.632.
- Holdout 97.683.

The average gap is now about 2.32 units Primary and 2.29 units Holdout.

PEAK BANKROLL AND MULTIPLIERS
Maximum Selective-Aggression bankroll:
- Primary: 126.5 at Hand 26.
- Holdout: 124.5 at Hand 15.

No session finishes at 2x, 3x, 5x or 10x.

DO DIMINISHING RETURNS HURT THE LATER MULTIPLIER PROGRAMME?
No. They answer a different question.

If later Phase 3.xx interventions produce small, zero or even negative incremental gains, that would tell us that the decision-policy optimisation has reached or crossed its useful boundary under the frozen 1-unit wagering architecture.

The later multiplier programme should NOT depend on every possible Phase 3.xx intervention being positive. It should inherit the BEST VALIDATED policy checkpoint, not blindly inherit the numerically latest phase.

For example, if a future 3.xx experiment worsened Primary and Holdout, we would preserve that result as evidence but reject that policy change. The multiplier programme would start from the last policy that remained convincingly validated.

This actually protects the multiplier research. It prevents us from carrying a harmful late-stage rule into the wager-optimisation experiments simply because it was tested later.

WHAT NEGATIVE RETURNS WOULD MEAN
A negative experimental result would not mean the overall hybrid is disadvantaged. It would mean:
'This particular additional override should not be included.'

The frozen policy checkpoint used for multiplier research can therefore be selected from the strongest validated 3.xx architecture after the policy programme is complete.

WHAT REMAINS AFTER 3.16
The remaining ranking is increasingly composed of narrower states:
- HARD 5/6 vs weak dealers;
- split-child HARD 12 and HARD 16;
- ordinary HARD 16 vs strong dealers;
- some HARD 9 split-child states.

This reinforces the late-stage picture: broad errors are becoming less dominant and residual damage is increasingly specialised.

NEXT QUESTION — PHASE 3.17
Run another aggregate family audit. Particular attention should be paid to the surprisingly persistent HARD 16 strong-dealer states and their split-child equivalents, while checking whether they are genuinely outside the earlier HARD 16 protection because of decision-stage/child-state architecture.

Do not alter them until that architectural relationship is explicitly verified.

PUBLICATION NOTE
This Layman Results report contains the complete explanatory interpretation. Other package files are technical evidence and audit material.
