name	bench
description	Run emitrix performance checks — throughput benchmark and stress suite — and judge regressions against recorded baselines. Use for performance questions and before merging changes to dispatch, ListenerStore, or the middleware chain.

Benchmarking emitrix

npm run build (benchmark runs against dist/)
node benchmarks/loadTest.js — 1M awaited emits across 4 listeners (incl. one wildcard)
npm run test:stress — memory soak with leak bounds, seeded fuzz, concurrency storms

Benchmark loop: ~700 ms for 1M awaited emits
~600 ns/emit with envelope + middleware + 4 mixed listeners (~1.6M emits/s/core)
Plain emit, 4 exact listeners: ~570 ns; parallel dispatch, 8 async listeners: ~840 ns
Subscribe+unsubscribe churn: ~300 ns with 100 standing listeners, ~1.3 µs at 1k, ~11.7 µs at 10k — linear O(n) growth is expected and accepted (emit-heavy workload trade-off)
10k concurrent waitFor resolved by one emit: ~75 ms total

2x slower on the benchmark loop: look at ListenerStore.collect (the wildcard path concat+sort) and the middleware reduceRight chain first.
Soak-test heap growth over its 10 MB bound: something is retaining per-subscription state — check timer sets and detachSignal cleanup.
Fuzz failures print their seed — rerun is deterministic; do not shrug them off as flakes, they never are.

name	bench
description	Run emitrix performance checks — throughput benchmark and stress suite — and judge regressions against recorded baselines. Use for performance questions and before merging changes to dispatch, ListenerStore, or the middleware chain.

npm run build (benchmark runs against dist/)
node benchmarks/loadTest.js — 1M awaited emits across 4 listeners (incl. one wildcard)
npm run test:stress — memory soak with leak bounds, seeded fuzz, concurrency storms

Benchmark loop: ~700 ms for 1M awaited emits
~600 ns/emit with envelope + middleware + 4 mixed listeners (~1.6M emits/s/core)
Plain emit, 4 exact listeners: ~570 ns; parallel dispatch, 8 async listeners: ~840 ns
Subscribe+unsubscribe churn: ~300 ns with 100 standing listeners, ~1.3 µs at 1k, ~11.7 µs at 10k — linear O(n) growth is expected and accepted (emit-heavy workload trade-off)
10k concurrent waitFor resolved by one emit: ~75 ms total

2x slower on the benchmark loop: look at ListenerStore.collect (the wildcard path concat+sort) and the middleware reduceRight chain first.
Soak-test heap growth over its 10 MB bound: something is retaining per-subscription state — check timer sets and detachSignal cleanup.
Fuzz failures print their seed — rerun is deterministic; do not shrug them off as flakes, they never are.

bench