If cause and effect land within roughly 100ms, the brain naturally attributes the sound to its own action (sensorimotor synchrony). Past that window, a delayed sound is still heard — but it no longer feels bound to the tap. It just reads as "the screen feels slow."
For interaction sounds, latency is felt performance, so they should play immediately on local recognition of the gesture — not after a network round-trip finishes. Actions that need server confirmation (payment, submit) can still give a short "acknowledged" sound optimistically, then correct with a distinct sound and visual alert if it later fails. On low-end devices where frames can drop, check that the visual change itself isn't lagging before worrying about the sound.
Applies to button-tap feedback, keystroke sounds, and hit-confirmation sounds in games — anywhere "right now" is the whole point.
When to use
Play interaction sounds immediately, locally — don't wait on a server round-trip. Past 100ms, cause and effect stop feeling bound together.