How this catalog is built

Where a price comes from

A price here is a manufacturer's MSRP. It has two permitted sources: the manufacturer's own store, or Amazon's struck-through List Price, which Amazon requires the manufacturer to substantiate. That makes it the same kind of number as an MSRP rather than a report of what something sold for last week.

A current selling price is not a permitted source. Neither is a marketplace reseller's price or a figure scraped off a search-results page. Refusing them costs coverage, and it is worth it: mixing today's discount into the field that holds MSRPs makes sorting by price straddle two different meanings, and a discount goes stale with nothing in the build able to notice.

Amazon also shows a struck-through Typical price, which looks identical and is not the same thing. It is Amazon's own average of recent selling prices, substantiated by nobody. We read the label rather than the strikethrough, and a Typical price counts as no price at all.

Every pack size of one product is sourced from the same place on the same day. Filling the 1-pack from the manufacturer's store and leaving the 3-pack on an older Amazon figure would reproduce that straddle inside a single product page.

112 prices across 77 currently-sold routers, one per pack size.

What "we don't know" looks like here

This is the rule that shapes the site most, so it gets the most room.

The catalog has a vocabulary of 50 features, and each router's record lists the ones its manufacturer states. That record does not say which features a manufacturer left out, and those are different facts. A spec sheet that never mentions MU-MIMO is evidence about the spec sheet, not about the radio.

So a feature nobody has verified renders as —, and never as No. Printing No there would publish a claim about shipping hardware that nobody checked, on a page a reader is using to rule a router out. The blanks are common enough that the difference is not academic, and the table carries a legend saying so wherever one appears.

A real No requires a person. A comparison page can declare a feature genuinely absent from one of its routers, and the build rejects that declaration if it names an unknown feature, a router outside the comparison, or something the catalog already says the product has. The strong claim only ever appears where somebody made it.

Across the 15 published head-to-head pages, the feature rows hold 498 cells that say yes and 91 that say we don't know. 5 cells say no.

The prose is held to the same line. A writer here may not turn a blank cell into "the X lacks Y". Either the sentence says the manufacturer does not list it, or somebody verifies it and records the negative.

The OpenWrt support ledger is this rule carried all the way to a page of its own. All 79 published routers have been looked up in OpenWrt's own hardware table, and what got published is everything the lookup found: the 11 that matched, each with the model string, the release and the wireless driver behind it; the 67 where the answer was no, dated, because a negative there ages in one direction only; and the one nobody was willing to call.

Routers we are holding back

A router with no verifiable price does not get published with a guessed one. It stays in the catalog source as a draft, excluded from the listing, from search, from every guide, and from every comparison, until somebody finds a price that meets the rule above. A draft is waiting on a lookup, not on an opinion.

5 routers are held back today, across 7 unverified price points between them.

Discontinued routers

Discontinued models stay listed. People still own them, still shop them used, and still want to know what they bought. They are also the one case where a price can be genuinely unobtainable rather than merely un-looked-up, because the manufacturer has stopped publishing one and the surviving offers are liquidation.

So they publish with no price rather than with a fabricated one, and the absence is recorded as nothing rather than as zero. Price filters exclude them instead of matching them at the bottom, and both price sorts pin them last in either direction. A router with no price is not a cheap router.

2 of the 79 published routers are discontinued.

Photos

Product photos are copied into this site rather than linked from the manufacturer's server. A hotlinked image rots silently: the URL stops resolving, the page renders a broken frame, and nothing in the build has any way to know. Each photo we hold carries the URL it came from, the date we fetched it, and the credit that renders under it on the product page.

A missing photo is reported to us but never blocks publication, and that asymmetry is deliberate. A price is a claim, so a bad one makes the page false. A photo is identification, so its absence makes nothing on the page untrue. Where a manufacturer publishes only lifestyle shots and no clean product render, the card shows a placeholder. We do not substitute a box shot or a photo of a different pack size.

63 of 79 published routers have a sourced photo.

The numbers we refuse to inflate

A Wi-Fi router's headline badge adds up every radio in the box, including any radio reserved for the mesh backhaul that no client is ever allowed to join. It is a real number and it is not a speed anyone reaches. We record it, because reconciling against it is what catches a mis-transcribed radio table, and then we rank and headline two different figures: the bandwidth actually available to clients, and the fastest single radio a device can join, since a device joins one radio at a time.

The Netgear Orbi 970 Series wears a BE27000 badge and sums to 27,089 Mbps. Available to clients: 18,442 Mbps. Fastest single radio: 11,530 Mbps.

Comparisons apply a second restraint. Two radios whose rated speeds differ by a fraction of a percent are the same radio wearing different marketing. Calling that a difference hands one router a win, weighted on screen exactly like the port and feature gaps that actually separate them. So a rate gap below 2% is not treated as a difference. The displayed numbers are never altered. Both print exactly as the manufacturer states them. We only decline to call the gap between them meaningful.

10 rows across the published comparisons hold unequal numbers the site declines to call a difference. The closest is a "5 GHz" row of 2,880 against 2,882 Mbps, a gap of under 0.1%.

No "Best Overall" badges

Nothing on this site is awarded a win. On a comparison, the only mark is on a row where more of the thing is unambiguously more, and it marks the higher number rather than the better router. A price is never marked, because cheaper is not more. Which price is the right one is a judgement about your budget, and the table does not have your budget.

The mark also disappears the moment any router in the row has no value at all. A discontinued router with no price must not come out ahead on a row it cannot answer.

Retailer links

Where we know a retailer's identifier for a specific pack size, the product page links to it. Those links carry a disclosure sitting next to them, and it says they may be affiliate links, which is exact rather than coy. Neither Amazon nor Best Buy links carry an affiliate tag today, so every outbound link on this site is currently a plain link that earns nothing. An Amazon link with no tag is a handled state rather than a broken one: the link still works, and the click is logged as untagged rather than dropping the reader at an error to protect a commission that was never going to arrive.

What a commission cannot do is change what the catalog says. Whether a router has a retailer identifier is not an input to anything: not to which routers are listed, not to any filter, not to any of the sort orders, not to a spec, not to a price, and not to whether a router is published at all. A missing identifier is normal rather than an incomplete record, and the router publishes and ranks exactly as it would with one.

45 of 79 published routers carry at least one retailer identifier. The other 34 link nowhere rather than to a guess, and for 4 brands that is true of every router we list.

How a change gets reviewed

The catalog is not a database. Every router is a plain text file, and so is every guide, comparison, brand page and glossary entry, so every change to a spec is reviewed as a diff before it ships. That is the point rather than an implementation detail: the individual numbers are the product, most of them start as an AI extraction from a manufacturer's spec sheet, and a diff is the only review surface that shows a changed number as a changed number. A row quietly updated in a database shows nothing at all.

The schema is enforced at build time, not at review time. A radio table whose parts do not sum to the stated total fails the build. So does a WAN speed faster than any port the router declares, a comparison naming a router that does not exist, a glossary entry calling a spec standard when fewer than 40% of the routers here list it, and a photo declared in a file that is not actually present. A spec that is merely wrong still needs a human. A spec that is incoherent cannot ship.

Editorial pages have thresholds too, for the same reason. A guide needs at least 5 matching routers before it publishes and a brand page needs at least 2, so a page that would be a thin filtered copy of the listing waits for inventory instead of shipping as filler. 9 guides and 9 brand pages meet that bar today.

What it takes to change a record

Numbers here are wrong sometimes, and the honest position is that some of them are wrong right now. What fixes one is a manufacturer page that contradicts us, named field by field. A review, a retailer listing or a spec-aggregator entry is not enough on its own, for the same reason a selling price is not enough to set a price: none of them is the party that has to stand behind the number.

The same standard governs a correction to this page. It describes rules the code enforces, and if a rule here has drifted from the code, the code is what is true.