On the Dot

Sense of time

Stopping a hidden clock at 7.24 seconds relies on interval timing, the ability to judge durations from a fraction of a second up to minutes. Psychologists have measured it since the 1860s. This page summarises what they found and compares it with 61,936 rounds of On the Dot.

Updated 6 October 2026. Sources are listed at the end.

Production

Timing research uses four classic tasks. In verbal estimation, people say how long something lasted. In reproduction, they copy a duration they have just experienced. In comparison, they judge which of two intervals was longer. In production, the experimenter states a target in seconds and the participant marks off an interval they judge to match it (Grondin, 2010).

On the Dot is a production task. The only time cue it gives is the 1.5 seconds of visible clock at the start of each round, and every round ends with feedback: the time you stopped at and how far off it was.

Errors grow with the interval

The most consistent finding in timing research is that errors grow with the length of the interval. Gibbon's scalar expectancy theory, built on animal experiments, states it precisely: both the average and the spread of timed responses scale with the interval, so the spread divided by the average, called the coefficient of variation, stays constant (Gibbon, 1977). It is Weber's law applied to time.

People mostly fit the same pattern. In a review of human timing, Wearden and Lejeune (2008) found that much of the data shows accurate averages and a spread in proportion to the interval, with coefficients of variation often between 0.10 and 0.15, about half those seen in animals. Violations are common in classic production, reproduction and estimation tasks, but production with feedback after each trial fitted the pattern.

On the Dot gives feedback after every round, and its data follows the pattern. From round 2 on, the typical miss grows from 0.45 seconds on 5-second targets to 1.21 seconds on 14-second targets, but stays between 7.7% and 8.8% of the target at every length. For a bell-shaped spread of misses, a typical miss of 8% corresponds to a coefficient of variation of about 0.12. Player data has the chart. It is also why the game scores each round as a percentage of its target: on that scale a 14-second round is no harder than a 5-second one.

Vierordt's law

In 1868 the German physiologist Karl von Vierordt published experiments in which he and his colleagues reproduced durations. Short durations came out too long and long durations too short, with an indifference point in between where reproductions were accurate (Lejeune & Wearden, 2009). The pattern is now called Vierordt's law.

Later work showed that it depends on context. When Jazayeri and Shadlen (2010) had people reproduce intervals drawn from one of three ranges, the reproductions were pulled toward the middle of whichever range was in use, and the pull was strongest for the longest intervals. A Bayesian model that combines a noisy sense of the interval with knowledge of the range fitted every participant. Glasauer and Shi (2021) found that the same durations produced less pull toward the middle when they changed slowly from trial to trial than when they came in random order.

On the Dot draws its targets at random from 5 to 15 seconds, and its data shows a small pull of the same kind. From round 2 on, the median stop on 6-second targets comes 1.0% late and on 13- and 14-second targets 1.3% early, while 8- to 10-second targets, near the middle of the range, come out on time. The effect is about a hundredth of the target, and Vierordt's and Jazayeri's experiments were reproductions rather than productions, so the comparison is a loose one.

Counting

Counting changes the picture. In comparison tasks, Grondin, Meilleur-Wells and Lachance (1999) found that counting gave no benefit for intervals around one second but improved performance markedly at 2.5 seconds, and placed the point where counting starts to help at about 1.2 seconds. With intervals up to 4 seconds, counting slowed the growth of variability without stopping it (Grondin, Ouellet & Roussel, 2004).

Hinton and Rao (2004) trained people on targets of 8, 16 and 24 seconds. Without counting, the coefficient of variation stayed at about 0.15 at every length. Counting silently left the averages about as accurate, tightened every response, and made the coefficient of variation fall as the targets got longer, from 0.09 at 8 seconds to 0.06 at 24 seconds in their first experiment. Errors still grew with the interval, only more slowly than the interval itself. Grondin and Killeen (2009) found that counting or singing cut the spread of 6- to 24-second reproductions to about a third of what non-musicians managed without a strategy.

On the Dot never uses whole or quarter-second targets, so counting can bring a player close but never all the way there. The game's typical miss stays near 8% of the target on long and short rounds alike, which matches the pattern Hinton and Rao saw without counting. Their results suggest that careful counting would pay off most on the longest targets.

Clock speed

Many timing models describe an internal pacemaker whose pulses are counted. Anything that speeds up the pacemaker ends a produced interval early, because the count reaches its target sooner. Across four experiments by Penton-Voak and colleagues (1996), trains of clicks played before a timing judgement sped up the internal clock by about 10%. In the experiment that asked people to produce intervals of 0.5 to 0.9 seconds, five seconds of clicks made their productions shorter.

Raised body temperature almost always speeds up subjective time (Wearden & Penton-Voak, 1995). Emotional arousal generally does too, but the effect depends on the emotion: in studies with emotional faces, disgust caused no distortion and shame caused underestimation (Droit-Volet & Gil, 2009). Most of this evidence comes from intervals under two seconds.

Attention

Attention pushes the other way. In the attentional-gate model, pacemaker pulses reach the counter through a gate that opens wider the more attention goes to time (Zakay & Block, 1997). When attention is divided, fewer pulses get through, so more real time passes before the count reaches its target and productions run long.

Experiments bear this out. Producing 2- or 5-second intervals while tracking a target, searching a display or doing mental arithmetic made productions longer, more variable or both, and harder second tasks disrupted timing more (Brown, 1997). A meta-analysis of 117 experiments found production to be the timing task most affected by mental load (Block, Hancock & Zakay, 2010).

Feedback and practice

Feedback after each attempt helps, mainly by correcting a consistent lean toward early or late. With intervals of 4 to 12 seconds, feedback had substantial effects on production and small effects on reproduction, and seemed to work by adjusting the remembered target (Franssen & Vandierendonck, 2002). Accurate feedback centred people's responses on the target without shrinking their trial-to-trial scatter (Ryan & Robey, 2002).

The game's data fits the first finding. 65.9% of first rounds end early. After one round of feedback the split is close to even, between 50.3% and 51.5% early in rounds 2 to 5. Median accuracy keeps rising to round 5, but players who quit drop out of the later rounds, so part of that rise reflects who is still playing.

In the brain

Timing in the range of seconds to minutes relies on circuits linking the thalamus, the cortex and the striatum, part of the basal ganglia. The cerebellum mainly handles millisecond timing of movements. Drugs that act on dopamine affect the clock itself, while drugs that act on acetylcholine affect memory for durations, and timing is impaired in conditions that disrupt dopamine, including Parkinson's disease, Huntington's disease and schizophrenia (Buhusi & Meck, 2005).

One leading account, the striatal beat-frequency model, proposes that neurons in the striatum detect moments when many cortical oscillators fire together, and it reproduces the scalar property (Matell & Meck, 2004; Buhusi & Meck, 2005).

Musicians

Trained musicians time better than non-musicians on some tasks. In reproductions of 6 to 24 seconds, musicians had about half the coefficient of variation of non-musicians, and their relative error fell on longer targets (Grondin & Killeen, 2009). A comparison of 36 musicians and 36 non-musicians found the advantage on immediate listening tasks, such as rhythm perception and comparing two durations, but not on tasks that rely on a remembered standard (Rammsayer & Altenmüller, 2006).

Tips turns these findings into ways to score higher.

Play today's game

Sources

Game figures are from 61,936 rounds played between 2 August and 5 October 2026. Player data explains how they were counted.