Battery percentage is an estimate, and a fairly rough one

This is written to be used rather than admired. Each section below is a decision about battery percentage, and each one has a default.

There is no fuel gauge

A battery has no way to report remaining charge directly, so devices infer it from terminal voltage, current flow, temperature and a model of the cell. This is called coulomb counting when it integrates current over time, and it accumulates error that has to be corrected periodically. The correction happens at the extremes, which is why a full charge and a full discharge improve the estimate.

The number on screen is therefore a model output, and treating it as a measurement leads to a lot of confusion.

The discharge curve is the problem

Lithium-ion voltage falls steeply at the top and bottom of the range and is nearly flat through the middle. A flat curve means voltage tells you very little about charge in the range where you spend most of your time.

At the protocol level, that is why the middle of the battery seems to drain unevenly and why the last ten per cent behaves strangely. Different lithium chemistries have differently shaped curves, and the flatter the curve the better the cell holds its working voltage and the harder it is to read a state of charge from it.

Cold makes it worse immediately

Low temperature raises internal resistance, so voltage sags under load and the device concludes the battery is nearly empty. A phone that shuts down at twenty per cent outdoors in winter and restarts indoors was not lying in either direction. The charge was there; the ability to deliver it was temporarily not.

At the protocol level, charging a cold cell is the case that does permanent damage, because lithium plates onto the anode instead of moving into it, which is why devices refuse to charge below a threshold and warm themselves first.

Ageing changes the model, not the labelling

As a cell ages its actual capacity falls, but the device still reports a percentage of current full capacity. A hundred per cent on a three-year-old battery is a smaller quantity of energy than it was, which is why runtime falls while the number still reaches a hundred. Battery health readouts report the ratio of current to original capacity and are themselves estimates.

Those readouts are recalculated from complete cycles and reported in coarse steps, so a figure can sit unchanged for months and then fall by several points in a day without anything having happened suddenly.

What actually extends life

Avoiding sustained time at very high and very low charge, and avoiding heat, are the two interventions with clear evidence. Charge limits that stop at eighty per cent target exactly this and measurably slow degradation.

In practice, heat is the most damaging single factor, which is why charging inside a case in direct sun is worse than any charging speed. Long-term storage is the case people get wrong: a device put away full and left in a warm cupboard ages faster than one used daily, and roughly half charge somewhere cool is the usual recommendation.

Firmware updates change this behaviour more often than hardware does.

Why the number jumps

Estimators smooth their output deliberately, holding a percentage steady rather than letting it flicker with every change in load. When the underlying estimate has drifted too far to hide, the correction lands all at once, which reads as a fault and is the model catching up with itself.

A device that only ever cycles between forty and sixty per cent never visits the steep parts of the curve, so it collects no fresh evidence about where its endpoints are and the estimate slowly loses calibration. Firmware updates sometimes discard the stored model entirely, which is why the reading can behave oddly for a few days after an update and then settle without intervention.

The takeaway

The percentage is a model. Heat is the thing actually shortening the battery's life.

The constraint is almost always physical, and marketing rarely mentions which one.