Every step you take gives your body a small vertical jolt — heel hits ground, mass decelerates, everything you're carrying bounces a few millimetres. Count the jolts, count the steps. That's the whole idea, and it hasn't changed since watchmakers were building mechanical pedometers with swinging levers for 18th-century aristocrats who wanted to know how far they'd strolled.
What changed is the sensor. The lever became a tiny silicon accelerometer — a microscopic mass on springs, whose position is read electrically thousands of times a second. Your phone has one; that's how it knows you rotated it. And it's how a web page can count your steps with no hardware you don't already own.
1 – Measure total acceleration. The sensor reports three axes; combining them into one magnitude means the phone's orientation stops mattering — pocket, hand, upside down, all the same.
2 – Subtract gravity. Standing still, the sensor still reads 9.8 m/s². A slow running average tracks that baseline so only the changes — your movements — remain.
3 – Find the peaks. Each step shows up as a bump of roughly 1–3 m/s² above the baseline. A peak detector marks them, ignoring anything implausibly fast (more than ~3.5 per second isn't walking, it's a washing machine).
4 – Demand rhythm. This is the move that separates decent counters from wishful ones. Real walking is periodic — peaks arrive evenly, around 1.5–2 per second. Good pedometers refuse to count until several evenly spaced peaks arrive in a row, then credit them retroactively. Our step counter waits for four. Pick the phone up, wave it, drop it in a bag — irregular spikes, no rhythm, no count.
Put a wrist tracker, a phone, and a hip pedometer on the same walk and you'll get three numbers, usually within a few percent. Each reads a different body part: wrists swing (and get pinned on treadmill rails — a whole story of its own), hips bounce cleanly, pockets ride along with the thigh. None of them is lying; they're sampling different echoes of the same footfall. The consolation: for tracking change — this week versus last week — any consistent counter works, because its bias cancels out.
And the dirty secret of every step counter, ours included: the first few steps of a walk go uncounted while the rhythm gate makes up its mind. That's not a bug. A counter that counts everything immediately is a counter that counts nonsense enthusiastically.