How WattMate's numbers are checked

Updated 6 November 2026 · Who measures this.

Rules published in advance

We checked WattMate's numbers ourselves on 6 Macs in our lab: the app against each Mac's own power sensor, and that sensor against a USB-C meter not made by Apple. Rules, thresholds and scripts were published before the first counted run.

On all 6 Macs, WattMate listed our test programs first. Its whole-Mac power comes from Apple's sensor, which stayed within about 0.3 W of the meter. 63 counted runs, 138.7 hours of measuring.

Results, data and scripts are in Version 6 of the record.

Results

A1 and A2 give a verdict for each Mac, and there is no single pass across all 6. A3 is a measurement, with no pass or fail.

A2, known loads: passed on 6 Macs of 6

The rule asks for at least 9 of 10 rounds. Every load ranked first in 10 of 10 on every Mac; on macOS 14.2.1 after a fix of the GPU load in the run script, with both series in the record.

Ranking first does not mean the row shows all the watts. With the disk load the row shows 0.14 to 0.30 of the Mac's rise; the rest stays in Screen and system.

A3, real apps: measured on 6 Macs of 6, no threshold

The share of the rise in the Mac's power that the app's own row carries:

n/a means the Mac's rise was below 0.5 W, too small for a share. On the 3 Macs with a backlight sensor, the backlight alone rose by 1.78 to 2.20 W with ChatGPT in front and with video.

A1, Apple's sensor against a meter: within about 0.3 W

This compares Apple's sensor with the meter, not WattMate. They differ by about 0.3 W at most. By the 2% rule set in advance that passed on 1 Mac of 6, because 0.3 W is more than 2% at low power: on 5 Macs at full brightness, and on the M1 also with 1 core busy. Which of the 2 is closer to the truth is not known: the meter's accuracy was not checked.

Each row takes the largest value of 2 runs on different days. A plus sign means Apple's sensor read higher than the meter.

Every table, by Mac and by run, is in Version 6 of the record.

The 3 checks

What was run

An attempt is every start of a check's script, as the preregistration defines it, finished or not; in A2 every round counts, so the 132 attempts are 63 runs. 10 of the 132 were excluded and replaced: 9 A2 rounds and 1 A3 run, each listed with its reason in the record. Trial runs do not count and are listed apart: 6 for A1, 5 for A2, 29 for A3.

Around these checks, our lab ran 404 test runs and 870 machine-hours of measuring on these Macs: the checks themselves (63 runs, 138.7 hours), their trial runs (40 runs, 19.3 hours), and the study of AI apps, instrument checks and nights (301 runs, 711.4 hours).

The 6 Macs

macOS versions are as of the day the rules were published. The 2 M1 Pro Macs share a chip and differ in macOS: macOS 14 is the oldest WattMate supports. Each Mac is one machine with its own macOS, so a result belongs to that Mac, and a difference between Macs is not put down to the chip. The Macs and who measures on them.

What the checks do not show

In the preregistration's own words: "None of the checks confirms an app's watts as such, and none tests the battery minutes." A3 describes how much WattMate attributes to an app; it does not check independently what the app really costs.

There is no reference instrument. The meter's maker publishes its resolution but not its accuracy, so A1 checks the agreement of 2 instruments from different makers, not the accuracy of either. A2 runs on battery, where the meter has nowhere to go, so it says whether WattMate's row agrees with Apple's sensor, not with physics.

Why it is set up this way

The developer makes and sells WattMate, so the checks measure our own product. That is why the rules were written before the data, the scripts are frozen by hash, a change after a crash is recorded as a new version, and the data of every counted run is published. Every start of a check's script counts as an attempt, a run stopped by hand is published with its reason, and the record keeps every version, so anyone can see whether the rules moved.

Questions people ask

Is WattMate accurate?

The whole Mac's power comes from Apple's own sensor; an app's row is an estimate built from the chip's counters, and How we count says where that estimate stops. On 6 Macs, Apple's sensor stayed within about 0.3 W of a meter not made by Apple, and WattMate listed our test programs first on all 6. What an app really costs, these checks do not measure.

How is WattMate tested?

3 checks on each of 6 Macs: Apple's adapter input sensor against a USB-C meter in the cable, 4 loads whose answer is known in advance, and 3 real apps against the rise of the whole Mac. Rules, thresholds and the number of runs were fixed and published before the first counted run.

Who runs the checks?

The developer, on Macs set up by hand for each run. The checks are not independent of us; what makes them checkable is that the rules came first and the data is published.

Where to check us

Results, data and scripts, Version 6: doi.org/10.5281/zenodo.24002711. The data holds every counted run with the readings of all instruments; WattMate's database and the window pictures are left out, because the database records every app on the Mac and the pictures show the app's window. The preregistration, with every version kept: doi.org/10.5281/zenodo.23697150. The study of AI apps on the same kind of Macs, with its data and scripts: doi.org/10.5281/zenodo.23391844 (the article). If a number on this site looks wrong, support@wattmateapp.com.