Latency bench

Measures the spread that survives calibration. A constant offset subtracts out and is harmless; jitter is what makes accuracy scoring dishonest. Every tap is timestamped by both methods at once, so the two are compared on identical taps.

Salsa Musicality Trainer · T001 + T002 · answers Q08 · nothing on this page moves on the beat, by design

Device

Run

Input

Tempo

Taps to keep

Output path

Put the sound somewhere you can hear the attack clearly, then start.

tap here

Kept0

Discarded0

Median A–

Median B–

Tap every click, on the click — not after it. The first four are discarded while you settle. Don't watch the screen.

Result

No run yet.

Method	Median offset	Jitter σ	IQR	Circular σ	Outliers

A — legacy, ctx.currentTime in handler B — corrected, event.timeStamp + getOutputTimestamp() each dot is one tap, centred on that method's own median · shaded band is ±30 ms

Runs this session

#	Input	Output	BPM	n	σ A	σ B	Reduction
None yet.

No telemetry sink exists yet (T006) — paste this somewhere until it does.