Relay Race

Five gates stand along a thirty-cell track, spaced unevenly, and a runner sets off from the left at a steady pace. Your one job is to press at the instant the runner is level with each gate — not before, not after. Five legs, and the runner is quicker on every one of them. Each mark is measured against the exact tick the runner crossed, and at the end the game tells you the thing you cannot feel while you are doing it: whether you lean early or late, by how many milliseconds, and how consistently.

It runs entirely on your device: 0 network requests, works offline. No ads, no account, nothing to buy, and no betting, money or prizes of any kind.

1. Focus the board — click it, or press Tab until it is focused.
2. Read the track before the runner sets off. The row of posts is the track. The tall posts are the gates and the short ticks are ordinary cells; the header above tells you which leg you are on, which gate is next, and how wide that gate's window is in milliseconds.
3. Watch the runner, not a rhythm. The arrow under the track is the runner and it moves one cell at a time, left to right. The gates are deliberately not evenly spaced, so counting will not save you — you have to read the distance still to go.
4. Mark the gate. Press Space or Enter at the moment the runner is level with the next gate. On a touchscreen use the Mark button; it goes through exactly the same path as the key.
5. Watch the pips. The row under the track is one pip per gate on this leg. A judged gate shows a tick if you were dead on, a left chevron if you were early, a right chevron if you were late, and a cross if the gate went by unmarked. The pending gate carries a focus ring.
6. Mind the five runners. A gate you never marked costs a runner, and so does a press that lands on empty track — including one after the last gate, while the runner is on the run-out. Lose all five and the race stops there.
7. Read the lean line at the end. It names your direction, gives your average distance from the gates in milliseconds, tells you how many of your marks landed on that same side, and prints your spread next to it.

Do most people press early or late when a moving marker reaches a line?

Most people press early, and — far more usefully — almost everybody presses consistently one way. The task in front of you is what the psychophysics literature calls prediction motion: you are not reacting to an event, you are extrapolating one. By the time you could react to the runner arriving it has already gone, because a simple visual reaction takes something like 200 to 250 milliseconds and the runner crosses a cell in far less than that on the later legs. So the only way to be on time is to launch the press before the arrival, timed off the runner's speed and the distance left. That extrapolation is where the bias lives. Estimates of arrival time from a moving object are generally too short — people tend to say "now" before "now" has happened, and the tendency grows with the speed of the object and shrinks with how long you have been watching it. Layered on top is the motor delay: the gap between deciding to press and the key actually going down, which people compensate for by launching early and habitually overshoot. Working the other way, a player who waits to see the crossing before acting will land consistently late, because that is a reaction rather than a prediction. Which of those two habits you have is stable enough to be worth knowing, which is why this game bothers to hand you your own number instead of telling you a general fact about people.

What counts as marking a gate here?

Every gate has an exact tick the runner crosses it, and a window either side of that tick. A press inside the window marks the gate; a press outside every window costs a runner, as does a gate you never marked at all. The window is not a number somebody picked. It is computed from the run-in — the clear track between this gate and the one before it — and it runs from about 83 milliseconds either side after a long approach to about 150 after a short one, with the figure for the gate you are on printed above the track. Inside that window there is a tighter band, roughly a third of it, where the tick symbol appears instead of a chevron and the score pays the most. The reason the window is derived from the run-in rather than fixed across the leg is that accuracy at this task improves with viewing time: a single tolerance would make the gates that happen to follow a short gap much harder than the ones that follow a long gap, and the layout rather than you would be deciding how hard your leg turned out to be.

Is there any money, betting or in-app purchase in this?

None, and there never will be. There is no currency, nothing to buy, no runners to refill for a fee, and nothing to win or lose but the race in front of you. Five legs and five runners is the whole economy.

Why are the gates spaced unevenly?

Because evenly spaced gates would quietly turn this into a metronome. Give anyone four gates at a fixed interval and after the second one they stop watching the runner and start counting, and from then on every mark describes their internal beat instead of how they read a moving object — which would make the early-or-late figure at the end a measurement of the wrong thing entirely. Irregular gaps, drawn from the seed, force each gate to be judged afresh from the distance still to run. The spacing is not arbitrary either: every gap is between three and five cells, and that guaranteed minimum is what the fairness argument below rests on.

Does it get harder, and how exactly?

Two ways, and both follow from one number rather than being stacked on by hand. The runner's pace quickens by a fixed step each leg, with a small seeded wobble that is deliberately too small to reorder the ramp, so every leg really is quicker than the one before it on every seed — roughly three cells a second on the first leg and about eight on the last. Everything else follows: a quicker runner gives you less viewing time per gate, less time to correct a launch you have already started, and less clear track between gates in which to recover from a mistake. The windows themselves stay honest in milliseconds, so the difficulty is entirely in the reading rather than in a tolerance quietly being tightened behind your back.

Can a leg ever be dealt that is impossible to run?

No, and that is guaranteed by how a leg is built rather than checked afterwards. The generator writes the crossing grid first — the runner leaves cell zero and advances one cell every fixed number of ticks, so the gate on cell twelve is crossed on the twelfth multiple of that number, exactly — which means the perfect run already exists before anything is judged: press on each of those ticks. The windows are then derived from that grid and clamped so that two neighbouring windows can never touch, which is arithmetic rather than a tuning exercise. So a press belongs to at most one gate, a press on the crossing tick is always accepted, and the tests drive that reference runner through the real game on thirty different seeds and require a clean sweep every time. The layout is built the same way: every gap is drawn inside a band the track is chosen to accommodate, so no gate can ever be dealt past the finish or too close to its neighbour to be judged.

Can I play it with a keyboard or a screen reader?

The keyboard is the primary control — Space and Enter both mark, and there is nothing a pointer can do here that a key cannot. For a screen reader the race is announced in words: the leg, the pace in cells a second, the width of the current window in milliseconds, how many cells of track the runner still has before the next gate, and every gate as it is judged, as "dead on the gate", "early", "late" or "missed". Nothing anywhere in the game is distinguished by colour alone, and the live region never counts down in ticks — it reports the distance, which is exactly what is on screen. Be honest with yourself about one real limit, though: on the faster legs the distance updates about eight times a second, which is quicker than most screen readers will speak, so the last legs lean on sight or on the shared sound cues. See the Limits section below.

Does it work offline?

Yes. Once the page has loaded once, the whole game is on your device — every leg is generated locally from a seed, timing is counted in simulation ticks rather than read from any clock but your own device's frame loop, and no server is involved in play at any point. Nothing about your session leaves the device.

Five legs, five gates each, twenty-five marks in a complete race. That is a small sample for a timing measurement: a laboratory version of this task would take dozens of trials at each speed, so treat one race as something interesting about your afternoon rather than a fact about you. The simulation runs at sixty ticks a second, so every number is quantised to about seventeen milliseconds — a reading of "62 ms early" is really "somewhere near 62". Absolute figures also carry your hardware: keyboard scan rate, screen refresh and touch latency all shift the whole distribution, and a touchscreen press is measurably slower to register than a key press, so the two are not comparable. The direction of your lean survives all of that far better than its size, which is why the end card leads with the direction. Accessibility has a real ceiling here too: the game is fully keyboard operable and fully announced, but tracking a runner that moves several cells a second is a visual judgement at heart, and the announced distance updates faster than speech on the last two legs.

Beat Keep is the closest neighbour and the useful contrast: it takes the moving thing away and asks you to hold a tempo from memory, so it measures your internal clock where this one measures how you read something you can see. Side Step also hands you back a measurement of yourself — there, the cost of a conflicting cue in your own reaction times. Tap Order times a visual search rather than an arrival, and Quick Count asks what you can take in without counting at all.

Bookmark this page (Ctrl+D, or ⌘D on Mac) or install the app — it works offline the next time you need it.