Function dispatch_audio_fill 

pub fn dispatch_audio_fill(
    world: &World,
    out: &mut [f32],
    sample_rate: u32,
    channels: u16,
)

Run every OnAudioFill in world once, mixing their samples into out.

out is interleaved frames and is zeroed first; each source adds its contribution in AudioFillOrder order — the order is the only thing the dispatcher states, so sources at the same order run in no stated order. Then the world’s GlobalAudioGain — one when it carries none — scales the sum, which clamps to the unit range. This is the whole of the audio thread’s per-chunk work, and a caller stepping a world inline calls it the same way. The gain is the finished buffer’s alone: every source ran before it, which is why a source that mirrors the output applies it itself.

A source must not change the world’s set of sources while it runs: it has a shared &World, so structural work goes through unlit_ecs::Scope::queue and is applied by the world’s own driver between fills.