Live Research

How Accurately Can You Stop a Stopwatch?

Try to stop the clock at exactly 10.00 seconds using nothing but your own sense of time. Every anonymized attempt feeds a live, public dataset — no sign-up, runs in your browser.

Loading live data…

00.000

Live distribution of timing errors

Every bar below is a 100-millisecond slice of how far real attempts landed from the 10.00-second target, centered on zero. The dashed line marks a perfect stop; bars to its right are late stops, bars to its left are early stops.

What this means

Humans don't perceive time like a clock does. Without a visible display, your internal sense of "ten seconds" drifts with attention, mood, counting strategy and even how the last few seconds of the attempt felt to you — so a spread of results, not a single fixed number, is the honest description of human timing accuracy. Published time-perception research generally finds that people are decent at short, trained intervals but get systematically less precise as the interval grows and as feedback is removed, which is consistent with what a "stop at 10 seconds, no display" task should produce. That said, the numbers on this page are our dataset — from visitors to this tool, not a controlled lab sample — so treat them as a live, informal measurement rather than a peer-reviewed result.

Methodology & reproduce this

What's measured
error_ms = stop_time − 10,000ms, captured with performance.now() from the moment you press Start to the moment you press Stop. A stop before 500ms is treated as a misclick and discarded client-side before it's ever sent.
Headline statistic
The median absolute error across all stored attempts, which is resistant to the occasional wild outlier in a way a plain average is not. It only appears once at least 1,000 attempts have been recorded; before that the page shows the running attempt count instead.
Underlying dataset
The full set also carries a trimmed mean and 10th/90th-percentile bounds, available in the raw export below — the chart and headline on this page use only the median and the histogram.
Anti-spam
Each submission carries a short-lived signed token fetched right before you click Stop, plus a per-visitor daily limit on how many attempts count — together these curb casual bulk spam. Because the headline is the median rather than the mean, it stays robust even if a few stray values slip through.

Download the raw dataset (CSV) — timestamp and timing error per attempt, nothing else. That export, together with the methodology above, is the reproducible artifact: compute the median absolute error, and for the trimmed mean drop the lowest and highest 10% of rows before averaging — no proprietary steps or hidden data.

Frequently asked questions

How accurately can humans stop a stopwatch at exactly 10 seconds?
It varies a lot from person to person and attempt to attempt. Rather than guess, this page runs the test live and shows the real, current numbers above — the headline only switches from a raw attempt count to a verified median once at least 1,000 attempts have been recorded, so the reported figure is never based on a handful of tries.
How is the timing error measured?
error_ms is the difference, in milliseconds, between the moment you clicked Stop and exactly 10,000 milliseconds after you clicked Start. It is measured with the browser's high-resolution performance.now() clock, not the coarser wall clock, so display refresh rate and animation timing don't distort the result. A positive number means you stopped late; a negative number means you stopped early.
Is my attempt anonymous?
Yes. Each submission sends only your timing error and a short-lived signed token that proves the attempt came from this page. No name, account or raw IP address is stored — the server keeps a one-way salted hash of your IP purely to rate-limit abuse, and the public dataset export contains only a timestamp and the timing error, nothing else.
Why does the verified average only appear after 1,000 attempts?
With only a few dozen samples, one or two outliers — a missed click, a distracted tap — can swing an average wildly in either direction. We wait for at least 1,000 attempts and then report the median, which is far more resistant to those outliers, alongside a trimmed mean in the underlying dataset.

This study runs on the same clock as our free online stopwatch — try it yourself, laps and all.