Sim ↔ Godot Parity Audit (B2)

Generated: 2026-07-10. Re-run python3 scripts/audit_hustler_damage.py and python3 scripts/verify_enemy_deck_parity.py to refresh.

Scope

Validate Jul 9 combat fidelity fixes (e8ea5c5 damage accounting, 8ac1b1f enemy AI) before any .tres balance changes.

Family-tier enemy decks

verify_enemy_deck_parity.py (healthcare Family strict set):

Enemy	Status
The Son (Junior Associate)	All 5 cards OK
Nurse Sister	Checked in strict set
Junior Dev / UX Designer (AI)	Checked in strict set

Son AI sanity: first card = Unchecked Math (10 dmg), not Server Overclock. Matches Enemy.cs GetCardScore + Aggressive personality.

Non-Family healthcare enemies have 16 card-stat mismatches (mostly dmg tuples in sim vs absent in .tres loader). These do not affect Family F1/F2 balance work; flag for a later card-loader pass.

Hustler damage accounting

audit_hustler_damage.py (10 sampled F1 trials, seed 7):

F1 win rate: 290/300 via full simulate_run (matches gold-standard report).
Block carry depletes per card during the player turn.
Enemy hits use remaining player.block after self-damage (Turn 1 example: 17 raw vs block 12 → 5 HP lost).
Kill turns use HP delta (no double-apply on Burnout lethal).
Ideal T1 lines differ from greedy in some hands (FDA-first vs Overclock-first) but all audited trials still win.

Jul 9 cliff — confirmed fidelity, not content nerfs

Snapshot	Pooled clear	Mean battles
2026-07-08	16.3%	~10.5
2026-07-09 post-fix	0.67%	~5.6

Git bisect (100 trials): e8ea5c5 → 1.2%, 8ac1b1f → 1.7%. Starter .tres nerfs are not the cause.

Open parity gaps (do not block A2 metric, track separately)

MidTier+ enemy card loader — 16 healthcare non-Family mismatches; sim uses hardcoded CARDS fallbacks.
Glitch vs block — playstyle audit notes opponent glitch may not reduce block in sim (pre-existing).
Manual Godot spot-check — recommended for Son T1/T2: confirm block-after-self-damage and Unchecked Math opener in live build.

Ruling

Proceed with sim-first balance on faithful model. Use A2 metric (ideal combat + greedy shop) as the tuning target. No .tres changes until ideal-combat clear rates are baselined and any open Godot discrepancies above are spot-checked.

Commands

python3 scripts/audit_hustler_damage.py
python3 scripts/verify_enemy_deck_parity.py
python3 scripts/run_simulator.py --sector healthcare --trials 100 --seed 7 --combat-policy ideal
python3 scripts/test_ideal_combat.py