Round 1 is the pre-rebalance main baseline. Round 2 is the feature/preset-deck-tuning branch after a full card-behavior audit — all 70 cards reconciled description-vs-implementation and the delivery bugs fixed. Every number here is the real Hazard-Pattern Combat engine, run headless through the project's own playtest harness at 200 seeded runs per cell. The rebalance genuinely lifted the weak early decks; the mid/late shortfall turned out to be structural, not a numbers problem.
These numbers are real. Every figure below is the actual Hazard-Pattern Combat engine, run headless through the project's own playtest harness (runPlaytestMatrix → simulateHazardPatternCombat) at 200 seeded runs per cell. Round 1 is the pre-rebalance main baseline; Round 2 is feature/preset-deck-tuning after a full card-behavior audit (all 70 cards reconciled description-vs-implementation, and the delivery bugs fixed). Each deck is its themed preset, played by the harness's scripted "blind" witness (drafts only on player-visible information — the realistic-player difficulty read). Paths scale player and enemy together: Early = L3 vs the coast's small griefs, Mid = L20 tier-2, Late = L45 tier-3 bosses (1,000–1,500 HP). Impossible was not run this pass.
The headline: this was a correctness pass, not a balance pass. The audit fixed cards that lied about what they did (a finisher that dealt zero damage, a curse dealing the wrong damage, a die silently lost) — but making cards honest barely moved win rates, because the mid/late shortfall is structural: affliction-payoff bursts cap at 80 HP against four-digit late-game pools, and durationed DoTs fall off before they erode a boss. The real balance lever is the deferred plan to re-home the themed afflictions as persistent (battle-long, no tick-down) disenchant curses.
Penitent regressed: mid fell 77.6% → 57.6%, late 4.6% → 0.7%. This is the one deck where the branch's rebalance measurably hurt — the audit made its cards honest but changed no balance. Fallen comes online as intended, yet the reworked payoff bursts still cap out against boss HP, so the net effect at mid/late was a loss, not a gain.
Grace is not a regression — it's the clearest structural wall. It sits unchanged (early 100%, mid 80%, late 0%). The root cause is in the engine: swayCapitulates() checks state.sway >= state.enemy.health — the enemy's current HP. Grace deals zero HP damage by design ("the deck that never strikes"), so the enemy's HP never moves; the capitulation bar stays pinned at full starting HP all fight. Against 1,000+ HP late bosses that target is simply unreachable, no matter how fast SWAY accrues.
Neither fix is a card change. Penitent and every DoT/affliction deck are gated by the shared 80-HP payoff-burst cap against four-digit boss pools. Grace needs the Dawncaster Charmed-style threshold — lose if HP < SWAY stack, reduced by damage — instead of "match their entire health bar." Both are the deferred persistent-curse / capitulation-rework lever, not per-card tuning.