The number is unreliable. The change isn't

Two phones side by side rarely show the same pressure. Yet they show exactly the same drop. The whole app rests on that difference.

Put your phone on the table next to someone else's, and ask them both what the air pressure is right now. The chance that they agree is small. Not because one sensor is faulty. Because that's how phone barometers behave — and it's the single most important technical fact in this whole collection of articles.

Two reasons the number differs

One reason is banal: height. A couple of floors up or down shifts the pressure measurably, and two phones are rarely at exactly the same height above sea level, even lying on the same table in the same room — not if the building isn't perfectly level, or if one phone is sitting in a bag a metre above the other.

The other reason is less obvious: sensor bias. Every tiny pressure sensor built has its own microscopic deviation from the factory — a touch too high or too low, consistently, every time. Two units of the same sensor model, in two units of the same phone model, will rarely show exactly the same number, even under identical conditions.

Meteorologists trying to fold pressure data from thousands of different phones into a single weather model have run into exactly this problem. Those two error sources — height and sensor bias — are precisely why researchers have had to build machine-learning methods to correct the collected pressure figures before they could be used in a forecast.

The error that disappears when you subtract

Here's the part that makes the sensor useful anyway. Sensor bias is a constant error. The same phone is consistently too high or too low — not randomly from reading to reading, but systematically, every time. And a constant error has a convenient property: it disappears if you subtract two readings from each other.

Say your phone's sensor consistently reads 3 hPa too high. If it measures 1013 hPa at 12 and 1010 hPa at 15, the true values are 1010 and 1007. The error is the same in both places — plus 3 — so the difference between the two readings, minus 3 hPa, is exactly the same whether you use the wrong numbers or the right ones. The bias goes out with the bathwater.

The same goes for the height error, as long as the phone itself doesn't change height between the two readings. That's why the research points to the trend specifically — whether pressure is rising or falling, and by how much — as extremely reliable, while the exact individual values are not.

Why it isn't enough to just look at yesterday

The point only holds as long as you compare a phone with itself. That's why Aneroid never tells you an absolute pressure level as some kind of official value, and never compares your phone with someone else's. The app only compares your pressure now with your own pressure three hours ago, on the same sensor. The same bias that made the number imprecise cancels out of the sum every time.

It's also why a flight, or a trip up a skyscraper, can throw off a reading — that changes the phone's height, not just the weather, and then the subtraction no longer delivers what it promises.