Skip to content

How we judge and rank

What a verdict is assembled from, which sources are given weight, how a shelf ends up in the order it is in, and the pressures this site is designed to resist.

A verdict counts for something only when its workings are visible. Here are ours.

What a verdict is assembled from

Every verdict on this site is assembled the same way, and each step leaves something on the page you can check for yourself: we read the specifications, the manuals and the compatibility lists a maker publishes, we set them against the standards those figures are defined by, we weigh what people who have run the hardware for years consistently report, and every figure printed here is named alongside the document it came from.

Most of the weight sits on that last rule. A figure carrying its source can be disputed; a figure without one is ornament. When a manual states one number and the sales page states another, the manual wins. When no trustworthy figure exists anywhere, the page records that fact and the field stays blank rather than being padded with something that merely reads well.

What counts, and in which order

Three questions set the order, and they are asked in this sequence: will it fit what you already own, will it do the job, and how many years will it go on doing it. In practice that comes down to six things:

  • The published standard that settles whether it fits at all. The length code on an M.2 drive and the standoff positions a slot offers, the PCIe generation and lane count a connector is wired for rather than shaped like, the PoE class a device asks of a switch, the category and shielding code of a cable against the panel it terminates into, the usable internal depth of a cabinet against the deepest thing going in it. This comes first because it is why most of this hardware goes back, and because a photograph answers none of it.
  • The governing figure, and what the maker counted to arrive at it. A class figure with a note of how many radios it adds together, a switching capacity that assumes every port saturated in both directions, a per-port PoE budget kept separate from the total budget shared across the unit, a runtime chart read at a named load in watts rather than at a headline, a sequential figure taken at a queue depth ordinary use never reaches. Where a maker publishes a headline without the condition that gives it meaning, we say so instead of repeating it.
  • The detail that decides whether it is still installed in year five. Whether a switch is fanless, whether an enclosure ships the thermal pad a fast drive needs to reach its shell, whether a mesh node accepts a wired backhaul, whether the memory in a NAS is socketed or soldered, whether a plug is made for the stranded conductors of a lead that will be moved. Every family has one such detail, and that is the one described here.
  • What the box does not contain, and what still has to be paid for. The PoE injector an access point is often sold without, the mains adapter a 3.5-inch enclosure cannot run without, the cage nuts a rack shelf may or may not arrive with, the transceivers and lead a fibre link needs at both ends, the drives a NAS enclosure is always sold empty of. This is weighted more heavily here than on most sites, deliberately: a purchase that needs a second purchase to work is not the price on the listing.
  • Spares, expansion and running cost. Whether a UPS takes a replaceable battery pack described by a cell reference anyone can match; whether a NAS can grow at all once its bays are full, given that adding a bay usually means another enclosure; whether a keystone panel accepts modules of a different category later; whether a crimping tool takes replaceable blades, since the trimmer dulls long before the jaws do. A product that looks slightly worse on the headline sometimes ranks above one that looks better, for exactly these reasons.
  • Whether anyone sells it in your country. Each reference is confirmed store by store, and every page names the markets holding a live offer. Hardware nobody stocks where you are is not a recommendation, however good the controller or the case.

Which sources are given weight

Three sources, weighted differently and always named. Manufacturer specifications, manuals and compatibility lists, cited as such rather than repeated as fact — they are dependable for dimensions, port counts, connector types and slot support, and much less dependable for anything written as a headline rate. Published ratings where they exist (a PoE class, a category, a PCIe generation and lane count, a TBW figure, a workload rating in terabytes per year, a stated load in kilograms), always flagged as figures to confirm on the current listing rather than as permanent guarantees. And the weight of long-term owner reports, which is the only practical source for the question that matters most: what is this like in year three, after a firmware update has changed the interface, after the fan bearing has aged in a room people sit in, and after the maker has quietly moved a range onto a different controller?

When a judgement rests on one of those three rather than all of them, the page names which. When a figure could not be sourced anywhere, the field stays empty and the gap is recorded as a gap. Nothing is ever completed by inference.

What an order on a shelf is telling you

We publish no scores. There is no number out of ten on this site, and there never has been — a single figure would have to weigh a tri-band mesh pack against a bag of RJ45 plugs, and it cannot. What gets published instead is an order: inside a family nothing is placed by accident, and the reason for each position is written in the text next to it.

Within a hub the order runs entry, mid, then top. A buyer arrives with a budget rather than with an ambition, and a page that opens on the ten-gigabit option answers a question nobody asked.

The pressures this site is designed to resist

Affiliate sites drift towards expensive products, towards whatever is newest, and towards whatever converts. Our defence is structural rather than a promise: no ranking on this site has ever been reordered because of what a product pays, several pages recommend a length of cable instead of a new radio, and no product has ever been placed in exchange for anything.

One distortion is peculiar to this subject, and saying it out loud beats letting it operate in the background. A large printed number sells hardware. A page written around the biggest figure on the shelf closes by recommending the quad-band system, and says nothing about the four gigabit sockets on the back that cap everything behind it. No page here is built that way. Where the honest answer is that an upgrade changes very little — a newer wireless generation for clients that cannot speak it, a managed switch in a house with nothing to segment, a five-gigabit adapter on a socket that cannot feed it — the page says so.

The one bias we cannot remove is availability. We cover what is genuinely purchasable across Amazon's US, UK, German, French, Italian and Spanish stores, which means excellent hardware sold only through specialist network distributors and business channels is under-represented here. Where that matters to the recommendation, the page says so.

Published figures are quoted here, not recalculated

This hardware is sold on figures, and those figures are the easiest thing on a listing to repeat carelessly. The rule that governs every page that touches them: every class figure, port speed, PoE class and power budget, PCIe generation and lane count, form factor code, capacity, sequential and random figure, TBW rating, MTBF statistic, VA and watt rating, dimension and stated load on this site is reported from the manufacturer's own documentation, with its unit written as published and its source named, and never recalculated or inferred at this end. Should a maker have published nothing, the page states that the figure is unpublished instead of offering one that would pass for it.

Capacity is where that rule does the most work. A maker counts a terabyte as a thousand billion bytes, so a 4 TB drive holds 4×10¹² of them and appears as roughly 3.64 TiB on a machine that counts in units of 1024. Nothing has gone missing: those are two conventions for one number of bytes. This site publishes the maker's figure in the maker's unit, says which convention it belongs to, and leaves the conversion to the operating system that performs it. And no product here is called faster, cooler or longer-lived than another on the strength of a specification sheet alone.

What "AX3000" and "10G" actually promise

One sentence governs the whole site, and it accounts for why several of these pages read more plainly than the ones they compete with: a published figure describes something; it promises nothing. A class figure adds the theoretical peak rates of separate radios together and describes no link at all; two units printing the same one can divide it very differently between bands. "10G" names a signalling class defined by a standard, not a rate a file copy will show, and the slowest element in a chain — a socket, a bridge chip, a lead, a mechanism — is what settles the outcome. A stated coverage area rests on each maker's own assumptions about walls and ceilings, and no two makers assume the same building. We write it that way because the alternative — treating a printed number as a promise — sells more hardware and leaves a reader disappointed by a purchase that did exactly what its specification said.

Two things this site will not say

Two boundaries here are not negotiable, and both are better said outright than left to be inferred. A RAID level is not a backup. Redundancy is a way of carrying on through a drive failure, and that is the whole of what it is: it does not survive a deletion, a ransom encryption, a controller fault that writes nonsense to every member, a theft or a fire, and an array rebuild is the moment a second drive is most likely to go. Every page that describes a redundancy level describes it in those terms, and says that a copy on separate hardware in another place is a different arrangement which nothing on the specification sheet replaces. And TBW, MTBF and a warranty term are three separate published quantities, none of which is a lifespan: the first states a quantity of writing the drive is rated for, the second is a statistic across a population of drives in hours, the third is a commercial term that usually ends at whichever limit arrives first. This site reports each as the declaration it is, and never blends them into a prediction about the unit you are holding.

The point at which a page here goes quiet

Nothing here replaces the manufacturer's own instructions. For anything that carries mains power, holds a sealed lead-acid battery, or has to be earthed to work as intended, the page points at them — and at the wiring rules that apply where you live — rather than paraphrasing them, even though a page that stops earns nothing.

One boundary is printed again on every page it touches rather than left to inference: a compatibility answer comes from a published standard or a published list, never from an impression. The slot table in a motherboard manual says how many lanes a connector is wired for; the switch maker's own module list says which transceiver coding it will accept; the printer list on a print server says whether your exact model is served and whether scanning is included; the cabinet drawing gives the usable depth between the rails. Where a maker publishes none of that, the page says the answer is unpublished rather than guessing at it from a photograph or a range name.

The questions that come up before an order

Can a placement or a review be bought here?

No. Not money, not a free unit, not an arrangement of any other sort has ever secured a position in a ranking on this site. If that ever changed, the page concerned would carry the fact in its own body text rather than in a line at the bottom of the site.

What does a verdict here actually stand on?

Manufacturer specifications, manuals and compatibility lists, the standards those figures are defined by, and the weight of what long-term owners consistently report — and each page names which of those a given figure came from. That is why a class figure or a TBW rating appears here as a published quantity, in the unit its maker used, rather than as a claim about what a link in your own building would carry.

How can a cheap unit sit above a costly one?

Because the order runs against what a product has to do, not against its price. A fanless five-port switch in a room where somebody works beats a managed unit whose fan never stops, and no amount of extra specification fixes the noise — so it is placed accordingly.

Does a page here ever end with "buy nothing"?

Yes. Several problems covered here are settled by wiring one mesh node back to the router instead of letting it relay over the air, by moving a powerline adapter off an extension lead and into the wall, or by fitting the thermal pad already in the box of an enclosure — rather than by a purchase, and we say so on the page even though it earns nothing.

Why do you never turn a class figure into a speed?

Because the two are different quantities. A figure such as AX3000 is the arithmetic sum of the theoretical peak rates the standard defines for each radio in the unit, computed from channel width and stream count; no client is connected to every radio at once, so nothing reaches the total. We publish the figure the maker printed, we say what it adds together, and we leave it at that.

Last reviewed 9 September 2026