iii · elébe — ahead of the beat

ütemező · rhythm — measuring how late you are

Tap with the click. Four beats count you in, then keep going. Almost everyone lands slightly early — typically twenty to fifty milliseconds ahead of a sound they believe they are landing on. It is called negative mean asynchrony and it has been in the literature since the 1890s. This piece measures yours, announces nothing, and then quietly stops keeping its own time and starts keeping yours: the click moves onto where you actually tap, and takes your tempo with it. The game is to notice. When you think it has happened, say so.

Every number here is contaminated, and by a known amount. The figure the panel calls corrected has your browser’s reported output latency subtracted from it. What cannot be subtracted is the delivery time of the keyboard or pointer event that stamped your tap, which no browser reports and which is worth several milliseconds with jitter of its own. So the mean is roughly right and the spread is inflated. On Bela, where the round trip is near 1 ms with sub-sample jitter, neither correction is needed — and that difference, not a preference for embedded hardware, is why the research builds do not run in a browser.

what this strand isolates

Sensorimotor synchronisation — the oldest and best-instrumented paradigm in experimental psychology’s dealings with music, and the strand whose data would be most immediately legible to researchers outside this project.

The source distinguishes two quantities with real precision: lag (késés, a systematic offset) and imprecision in time (időbeli pontatlanság, variance). They are separately meaningful and separately trainable. A player can be reliably 45 ms early with a standard deviation of 8 ms — a very good drummer with an idiosyncratic feel — or dead on average with a standard deviation of 60 ms, which is a different person entirely. The panel reports them apart for that reason.

It also splits the variance again. Wing and Kristofferson showed in 1973 that the variance of tap intervals decomposes into an internal timekeeper term and a peripheral motor term, and that the two can be separated from the lag-one autocovariance of the intervals alone — the motor variance is what makes successive intervals negatively correlated, because a tap that is late shortens the interval after it. That decomposition runs on data these games produce as a by-product of being played, so the panel prints it: a timekeeper figure and a motor figure, in milliseconds. Where the model does not fit — a positive autocovariance, which happens when someone is drifting rather than keeping time — it says so instead of printing a number.

Then it takes your beat away. Over thirty-two beats the click migrates onto your own taps and adopts your tempo, after which it keeps following you: rush deliberately and it will come with you, slowly. It is a rigorous protocol and a small piece of theatre at once. The instrument learns your error and adopts it as its truth, which is either the most flattering thing a machine has ever done for you or quietly sinister, depending on the minute. Then it perturbs itself by twenty-five milliseconds, four times, without telling you, to see how much of a displacement you correct on the very next tap — a number that is mostly not under your control at all.

the rest of the strand

1. Ü2 Dobpárbaj · Drum Duel
   Four buttons on a modelled kit — membrane, wood, metal, shaker — against a modelled partner or another player, in call and response with a closing window: four bars to answer, then two, then one, then a half. The complexity ladder runs exactly as the source proposes: up the syncopation index, then into metrical change (4/4 → 7/8 → 5/4 → additive Balkan groupings of 2+2+3), then into subdivision. The Hungarian and South Slav rhythmic repertoire is not decoration here; it is the most efficient source of asymmetric metres in Europe. It is the strand’s engine of progression, and the one place the source’s phrase about developing cognitive function through task complexity gets a falsifiable implementation: a ladder with numbered rungs and a ceiling that can be measured before and after. measures — ceiling metre reached · error taxonomy, phase / pattern / timbre · recovery time after a dropped bar · asymmetry between limbs
2. Ü3 Késés · The Lag
   Keep a groove going. Without warning the system inserts additional delay between press and sound: 0 → 8 → 16 → 30 → 60 ms, then back down. The player is told nothing. The game finds the point at which the instrument stops feeling like a limb and starts feeling like a message being sent somewhere. It turns the project’s own hardest engineering constraint into a playable instrument, and the thresholds it yields are publishable in their own right — in a literature whose canonical numbers rest on comparatively few participants, a game played by four thousand gallery visitors is a larger sample than the field has ever had. measures — per-player latency detection threshold · tolerance threshold, detected but still playable · musician against non-musician · dependence on timbre, tempo and instrument

Neither is built here, and Ü3 in particular cannot be: it needs a 1 ms floor so that the added latency is the only latency, and a browser starts around thirty times that. Running it here would measure the machine and call it the player.