Your Phone’s Battery Percentage Is an Estimate

Your phone says 8 per cent. You open the camera, record a short video, and the number falls to 5. A few minutes after you stop, it appears to steady. On another day, a cold phone switches off while the display still promises charge. The obvious conclusion is that the percentage is broken—or perhaps that the manufacturer is hiding something.

But the number was never a direct reading. There is no tiny dipstick inside a lithium-ion cell. The percentage is an estimate produced by a fuel-gauge system: sensors measure electrical behaviour, while an algorithm infers how much usable charge remains. That distinction matters because an estimate can change when the evidence changes, even though no energy has mysteriously returned.

The useful question is therefore not “Is the percentage true?” but “What is the gauge estimating, from which signals, and under what conditions does its confidence weaken?” Once we treat the number as a model rather than a fact, a jumping percentage becomes less mysterious—and sometimes diagnostically useful.

Counting what cannot be seen

One basic method is coulomb counting. A small sensor measures current flowing into and out of the battery; the gauge integrates that flow over time. If it begins with a reliable estimate and measures perfectly, it can keep a running account of charge used and charge added.

In practice, small measurement errors accumulate. The gauge also needs to know the denominator: the battery’s present usable capacity, not the capacity printed on the box when it was new. Those are different quantities. State of charge asks how full the battery is now; state of health asks how its current capacity and power capability compare with its earlier condition.

A second clue is voltage. A rested cell’s open-circuit voltage has a relationship with its state of charge, but a working phone is rarely at rest. Its voltage dips under load and then relaxes. Temperature and internal resistance change that response. Parts of a lithium-ion cell’s voltage curve are also flatter than others, so the same small voltage error can imply a larger uncertainty about charge.

Modern gauges combine these imperfect clues. Texas Instruments describes its Impedance Track approach as using depth of discharge, total chemical capacity, resistance, load and temperature, alongside coulomb counting. Analog Devices describes a comparable hybrid in which a battery model estimates open-circuit voltage during use and corrects the running count. These are not descriptions of every phone’s exact implementation, but they show the engineering pattern: count charge, model the cell, compare prediction with measured behaviour, then correct. The percentage is the output of that reconciliation.

Why the number can jump

Imagine a phone doing an easy task at room temperature. Current draw is modest, voltage is stable, and the gauge’s model agrees with its recent history. The estimate may look wonderfully smooth.

Now start a demanding game, video call or camera session. Current rises. Because every battery has internal resistance, terminal voltage falls under the heavier load. A cold or chemically aged cell tends to have higher effective impedance, so the dip can be larger. The gauge must decide how much charge remains and whether the cell can deliver enough instantaneous power before voltage reaches the device’s safe operating limit.

That is why remaining energy and remaining runtime are related but not identical. A phone can contain some charge yet be unable to supply a peak demand without its voltage falling too far. Apple’s public explanation of unexpected iPhone shutdowns describes exactly this interaction among chemical age, low charge, cold temperature, impedance and peak power. It also explains why the system may manage performance after a shutdown: the immediate problem can be power delivery, not an empty chemical reservoir.

When the load ends, voltage relaxes. The model receives a less distorted clue and may revise its estimate. A software restart can also cause the operating system and gauge to exchange fresh information. Neither process creates charge. The displayed number has moved because the inference has been updated.

This is also why ageing makes some gauges look less trustworthy. A battery that holds less charge and has higher resistance can move through the useful range faster, sag more under load and depart further from the model learned earlier in its life. As Alkemata has explored with e-bike batteries, a battery is not simply a tank. Its chemistry, temperature, history and power demand shape what “remaining” means.

A useful number, not a laboratory instrument

The strongest counterargument is practical: people do not want a confidence interval every time they glance at a screen. They need one legible number, and modern gauges are usually good enough to decide whether to leave home without a charger. A smoothly filtered percentage may be more useful than a twitching raw estimate.

That is right. The lesson is not to distrust the gauge or obsess over each percentage point. It is to match the precision of our decisions to the precision of the signal. At 70 per cent, a one-point error rarely matters. Near empty, in cold conditions, with an older battery and a high-power task ahead, the same apparent precision deserves less confidence.

Manufacturers also face a design trade-off. If software exposes every correction, users may interpret honest uncertainty as failure. If it hides corrections too aggressively, the final fall or shutdown feels deceptive. A better interface might pair the simple percentage with contextual evidence already available on many devices: battery health, recent drain by app, a service warning, or an explicit notice that cold conditions may reduce peak capability. Simplicity need not require false certainty.

Run a three-day estimate diary

You can examine your own gauge without deliberately draining, heating or chilling the phone. For three ordinary days, choose one repeatable ten-minute task—playing a downloaded video at the same brightness, for example. Begin once near 70 per cent, once near 40 per cent and once near 20 per cent, whenever those levels arise naturally.

Record four things: the starting and ending percentage; whether the phone felt cool, normal or warm; whether it was on Wi-Fi or mobile data; and the battery-health or service status shown by the operating system. Then leave the screen off for fifteen minutes and note whether the estimate changes after the phone rests. Do not force a deep discharge, place the phone in a refrigerator, or run stress-test software. The point is observation, not punishment.

A single irregular result proves little. Look for repetition. Does the same task consume far more displayed charge near empty? Do corrections appear mainly when the phone is cold, hot or under a heavy load? Does the device shut down with charge remaining, or show a service warning? Those patterns help separate ordinary estimation noise from a battery whose capacity or power capability may justify professional inspection.

The experiment also changes the relationship between user and device. Instead of treating the percentage as an oracle, you treat it as an instrument with a model, operating conditions and failure modes. The next useful step is modest: collect three comparable observations before changing settings or buying a replacement. A better decision begins not with a more confident number, but with a clearer account of how the number was made.