PHASE 3.03 — RUNNING TXT AUDIT RECORD

OBJECTIVE
Make the first controlled stabilisation intervention while preserving the Phase 3.02 frozen architecture: split-child HARD 20 STANDS rather than taking Maximum Aggression's automatic DOUBLE.

EVIDENCE HIERARCHY
1. Primary design authority: finite six-deck Phase 2.xx evidence and new Phase 3.xx data.
2. Earlier non-depleting/infinite-generator behaviour-profile data: historical corroboration and hypothesis generation only.
3. Old behaviour-profile evidence must not directly define a new policy unless finite-shoe evidence independently supports the same conclusion.

FROZEN VARIABLES
Six decks, deterministic session shuffle, 30 hands, 100-unit start, 1-unit base wager, dealer rules, blackjack payout, split limit, no in-session reshuffle, textbook control and Maximum Aggression policy are unchanged from Phase 2.02.

POLICY INFORMATION BOUNDARY
The future optimiser may use player hand composition/total, pair status, dealer upcard and legal actions. Phase 3.03 deliberately uses no previous-card-history, running count, remaining-rank inventory, hidden card or future-card feature. The new rule depends only on the current visible split-child HARD 20 state.

PHASE 3.03 DECISION RULE
ONE intervention only: if the hand is a split child, is hard (not soft), totals 20, and is at its first decision, Selective Aggression STANDS. Otherwise the frozen Maximum Aggression policy is unchanged. Textbook remains the control.
