How we test
Every measurement on this site was taken on the machines listed below, by hand, under conditions we state in the article. This page describes the method those articles refer back to.
The machines
MacBook Pro Apple M4 Pro
Why the rig is this small
One machine is a real constraint and we would rather state it than work around it. It means we do not cover hardware we have never touched, we do not compare six laptops we borrowed for a week, and we do not publish figures for a configuration we have not run. When an article needs a machine we do not have, the article does not get written. Every article currently on this site is about the Mac above, because that is the only device measurements have been taken on.
The upside of a small rig is that every number on this site comes from the same known baseline. When a figure changes between two articles, the change is in the software or the method, not in an unlisted difference between test units. The battery on this MacBook Pro has 190 cycles on it and that count is published, because a battery figure taken on a nearly new machine and one taken at 500 cycles are not the same measurement.
The method
Every number states its run count
A single reading is an anecdote. Anything we describe as measured was taken more than once, and the article says how many times. Where readings varied, the variance is published next to the figure rather than averaged away silently. Where a value can only be observed once, such as a build number or a cycle count, the run count is one and the article does not call it a measurement.
Conditions are stated before the result
Screen brightness, ambient temperature, power source, battery state, what else was running, and whether the machine had been idle. These change results more than most people expect, and an unstated condition is the most common way a benchmark becomes unreproducible. Timing runs are taken with the machine on AC unless the article is about battery, in which case the charge state at the start of the run is given.
The OS build is part of the result
Articles are stamped with the exact build string, currently macOS 26.3.1 (25D771280a), not just the marketing version. A procedure that works in one point release regularly stops working in the next, and a percentage that shifts after an update is a finding rather than an error. When we re-test, the stamp changes and the change is noted in the article.
Tools are the ones already on the machine
Measurements come from utilities that ship with macOS wherever possible:
ioreg, system_profiler, sysctl, vm_stat,
tmutil, mdutil, networkQuality and pmset.
Using stock tools means a reader can reproduce a result without installing anything and
without trusting a third-party utility's interpretation of the same underlying data. Where
a command is used, the article prints it in full so it can be pasted and run.
Raw values are preferred to summarised ones
Where the system exposes both a processed figure and the underlying reading, we publish both and show the arithmetic between them. The battery capacity figures are the clearest example: this machine reports a design capacity of 6249 mAh, a nominal charge capacity of 5869 mAh (93.92%) and a raw gauge capacity of 5717 mAh (91.49%). Publishing only the rounded percentage would hide a gap that is itself the interesting part.
What we will not do
- Publish a number we did not produce. No figures taken from a vendor page, a press release, another publication or a forum post are ever presented as our measurement. Where we cite an external figure it is attributed and marked as theirs.
- Estimate a result we could not run. If a test failed, timed out, or could not be completed on our hardware, the article says so instead of substituting a plausible value.
- Use images we did not make. Every screenshot is taken on the machines above. Diagrams are drawn by us. Nothing is lifted from documentation or from another site.
- Test devices we do not own. Coverage is limited by the rig, deliberately.
Limitations you should weigh
One machine is one sample. Silicon varies between units, thermal paste application varies, and a figure from this MacBook Pro is not a population average. Where a result would plausibly differ on another unit, the article says so. Ambient temperature in Hyderabad runs warmer than a climate-controlled lab for much of the year, which matters for any sustained-load measurement, so ambient temperature is recorded for thermal tests.
We are also not equipped for certain categories of measurement at all. There is no power meter on the wall, no external instrumentation, no anechoic chamber, no colorimeter. Anything requiring those is outside what this site can honestly claim, and we do not approximate it with software readings and present the result as equivalent.
Corrections
If a number here is wrong, we want to know, and the correction goes on the article with the date it was made rather than being edited in silently. Our approach to corrections is described in the editorial policy, and you can reach us through the contact page.