A modern iPhone carries three instruments that matter on a mountain: a GPS receiver, a barometric altimeter, and an inertial unit — accelerometer plus gyroscope. None of them was designed for skiing. Each one lies in its own characteristic way. But their lies are uncorrelated, and that’s the whole game: cross-check three flawed instruments and you get telemetry none could produce alone.
This is the long version of how Alpy turns those sensors into the numbers on your debrief — and why some of those numbers come with a tilde or a ± attached.
Satellite positioning solves your location from timing differences across visible satellites. Horizontally, with open sky, a phone sits within a few metres of the truth. Two things degrade on a mountain:
Altitude. Vertical error runs two to three times horizontal error by geometry alone — the satellites that would pin your height best are below the horizon. Add a valley wall blocking half the sky and ±15 m of altitude wobble is normal. For a ski app this is poison: altitude drives everything (run detection, vertical totals), and raw GPS altitude makes a stationary phone “climb” and “descend” all afternoon.
Speed. Phones don’t derive speed from successive positions (hopeless noise); they use the Doppler shift of the satellite carriers, which is much better — typically within 1–2 km/h in open terrain. But iOS reports a per-reading confidence alongside every speed, and that confidence collapses in trees and tight terrain. Most apps discard the confidence number. Alpy records the one attached to the moment your top speed was set, and prints the result as 54 ±3 km/h — because a headline number without its error bar is a guess wearing a suit.
The pressure sensor resolves altitude changes of well under a metre and reacts instantly — a sensor you could feel a staircase with. Its weaknesses mirror GPS’s strengths: it has no idea of your absolute altitude, and its baseline drifts as weather moves through. A front passing during your ski day can shift the pressure-derived altitude by tens of metres — ironic, since storm days are exactly when you’re skiing.
So the fusion is natural. Take the barometer’s beautiful short-term shape; anchor it to GPS’s absolute level with an offset learned slowly — over minutes, not seconds. Learned fast, the offset would re-import GPS wobble; learned slow, it absorbs weather drift while ignoring satellite noise. In our test harness the fused altitude tracks truth with less than a third of raw GPS’s error. Every point Alpy records while tracking rides the fused value, which is why run detection doesn’t stutter and vertical totals don’t inflate.
Raw accelerometer data is the noisiest signal on the phone — every chatter of the piste, every pocket bounce. It still yields two measurements nothing else can, both built on the same idea: look for signatures that can’t be faked by noise.
Airtime. In freefall, an accelerometer reads zero — you, the phone, and the sensor’s test mass all accelerate identically, so nothing presses on anything. It doesn’t matter which pocket the phone is in or which way it’s turned: total measured acceleration near 0 g is flight. Alpy calls it air when that signature holds for at least a quarter second (shorter dips are the suspension of your knees absorbing terrain), and discards “airs” longer than three seconds as sensor glitches, on the grounds that you did not, in fact, huck a seventy-metre cliff. An air that’s still open when tracking stops is thrown away too: unclosed air is unproven air.
Turn load. Subtract gravity’s direction and the acceleration left over, perpendicular to gravity, is the sideways push of your turns — carve hard and it climbs toward 1 g and beyond. The trap is spikes: a single landing rattle can read 3 g for one sample, and a “max G” feature that reports it is measuring your phone’s ride, not your skiing. Alpy smooths the lateral signal over roughly half a second and reports the peak of the smoothed curve — only a sustained carve can set it. The number still wears a tilde. It’s an estimate until we’ve validated it against instrumented skiing, and the app says so in the same breath it shows the number.
Turn counting uses GPS heading swings: linked turns sweep your course left-right-left, and each sustained swing past 30° (with hysteresis, so jitter can’t double-count) is a turn. The method’s blind spot is sampling: at one fix per second, tight slalom turns blur together and the count reads low. That’s why the debrief says “~14 turns,” never “14” — and why the trend matters more than the level. Cadence per run, plotted across the day, shows your rhythm fading as your legs go even if every individual count is a little short. With four or more comparable runs the app will put a number on the fade; with three it refuses, because three points claiming a trend is how instruments become horoscopes.
Every number in Alpy is classified: measured (GPS distance, fused vertical, speed with its ±), estimated (turns, airtime, turn load — always marked with ~), or sourced (weather, avalanche danger — always naming the provider). Nothing is invented, and when a sensor can’t support a claim, the claim doesn’t appear.
A phone will never match a lab. But a lab doesn’t fit in your jacket, and a phone that’s honest about its error bars beats a gadget that’s confident about numbers it can’t actually measure.
All of this telemetry is free in Alpy, with nothing leaving your phone. The app · More from the blog.