Unified delta benchmark

The unified delta workload tests a single logical observer reacting to added, changed, AND removed entities across three component types (State, Health, Mana) per tick. This is the shape Zay-ES's EntitySet is designed for; it exposes japes's per-component-type system registration overhead.

What it measures

Per-tick driver: 3 mutator systems iterate all entities and write 10% each (30% of entities touched total), 1% spawn, 1% despawn, 2% component strip-and-restore (Mana removed then re-added next tick). 3 observer systems count added, changed and removed events via multi-target @Filter. Zay-ES does mutations via EntitySet.applyChanges() on the same workload shape.

Results

	japes (6 systems)	Zay-ES (1 EntitySet)
10k entities	660 µs	1,067 µs
100k entities	4,854 µs	13,889 µs

japes beats Zay-ES at both 10k (1.62×) and 100k (2.72×). Both frameworks iterate all entities per component type, evaluate a game-logic condition (HP > 900, mana < 100, state % 10 == 0), and write only the ~10% that trigger. japes mutator systems read component values through tier-1 SoA primitive array loads; Zay-ES goes through data.getComponent() per entity (HashMap lookup). The SoA advantage grows with entity count because sequential array access scales better than per-entity HashMap probes.

What the japes code looks like

// Damage-over-time: entities above 900 HP take 1 damage per tick
@System
void damageOverTime(@Write Mut<Health> h) {
    if (h.get().hp() > 900) h.set(new Health(h.get().hp() - 1));
}

// Mana regeneration: entities below 100 mana regen 1 per tick
@System
void manaRegen(@Write Mut<Mana> m) {
    if (m.get().points() < 100) m.set(new Mana(m.get().points() + 1));
}

// Multi-target @Filter observers react to changes across 3 component types
@System
@Filter(value = Changed.class, target = {State.class, Health.class, Mana.class})
void observeChanges(@Read State s, @Read Health h, @Read Mana m) {
    counters.changed++;
}

Reproducing

# japes (multi-target @Filter, 5 systems)
./gradlew :benchmark:ecs-benchmark:jmhJar
java --enable-preview \
  -jar benchmark/ecs-benchmark/build/libs/ecs-benchmark-*-jmh.jar \
  "UnifiedDeltaBenchmark" \
  -p entityCount=10000,100000

# Zay-ES (1 EntitySet)
./gradlew :benchmark:ecs-benchmark-zayes:jmhJar
java -jar benchmark/ecs-benchmark-zayes/build/libs/ecs-benchmark-zayes-*-jmh.jar \
  "ZayEsUnifiedDeltaBenchmark" \
  -p entityCount=10000,100000