PHASE 3.08 — LAYMAN RESULTS EXPLANATION
SIXTH CONTROLLED STABILISATION INTERVENTION

WHAT CHANGED
Phase 3.08 keeps every protection validated in Phases 3.03–3.07.

It changes one coherent decision family only:
ANY first-decision HARD 17 now STANDS instead of Maximum Aggression automatically DOUBLING.

This applies whether HARD 17 is an ordinary hand or a split child. The action is not being invented for this experiment: STAND is the corresponding Textbook first action for the HARD 17 states that were dominating the remaining replicated evidence.

WHY WE MOVED FROM ONE ROW TO A FAMILY
Phase 3.07 gave an important warning. Its single split-child HARD 16 vs Ace correction was strongly damaging when encountered, but uncommon enough that correcting it improved average bankroll by only +0.021 units in each cohort.

After that intervention, HARD 17 did not appear as one isolated anomaly. Multiple HARD 17 states appeared prominently across dealer upcards and across ordinary/split-child contexts, with Textbook consistently prescribing STAND.

Phase 3.08 therefore tests one coherent HARD 17 family. This still changes only one decision principle, while addressing repeated exposure rather than chasing one rare row.

RESULT
Primary average ending bankroll rises from 88.705 in 3.07 to 90.594 in 3.08: +1.889 units.
Holdout rises from 88.482 to 90.478: +1.996 units.

Positive sessions rise from 12.72% to 16.14% Primary and from 12.18% to 16.14% Holdout.

This is a substantial recovery compared with the tiny 3.07 movement, and it replicates independently.

PEAK BANKROLL
Selective Aggression maximum observed bankroll:
- Primary: 130.0 units at Hand 30.
- Holdout: 132.0 units at Hand 30.

The Primary cohort maximum is lower than the 135-unit extreme observed in 3.07 even though average ending bankroll and positive-session frequency improve materially. This is useful evidence that optimisation of typical outcomes and optimisation of a single extreme peak are different objectives.

MULTIPLIER OUTCOMES
No session finished at 2x, 3x, 5x or 10x in either cohort. These remain observational measures only.

WHAT THIS TEACHES US
Phase 3.08 supports the lesson emerging from 3.07: intervention selection should not be based on per-encounter damage alone.

A coherent state family with repeated exposure can have much greater programme-level leverage than an individually severe but uncommon state. The research is therefore becoming better at distinguishing:
- severity when a bad decision occurs;
- frequency/exposure of that decision;
- aggregate effect on the 30-hand bankroll distribution.

WHAT REMAINS
The remaining replicated table now contains many HARD 16 states, alongside low-total HIT states. HARD 16 is especially important because the corresponding Textbook action is dealer-dependent: for example, some HARD 16 states HIT while others STAND. We must therefore NOT automatically create a blanket HARD 16 rule.

NEXT QUESTION
Phase 3.09 should analyse the HARD 16 evidence as a structured dealer-dependent family and determine whether a controlled intervention can preserve the correct Textbook action for each visible state without over-broadening the policy.
