How we rank

1. Repeated independent rankings per category

We never rank a category just once. Every refresh researches the board again, each time through a different lens, and the runs accumulate rather than replacing each other. The typical product on this site has been ranked in 6 separate runs.

The first three are the ones every refresh plays. They are deep research runs over the current reviews and buying guides of the most authoritative sources for that category (dedicated testing labs, specialist publications, well-regarded expert reviewers), carried out by an AI researcher with live web access. Community pulse works differently on purpose: instead of searching the web it is handed the actual recent discussion, Reddit, Hacker News and YouTube from the last 90 days ranked by real engagement, so those runs hear owners rather than reviewers. It runs where the discussion is deep enough to rank from, which is not every category.

Each run commits to its own strict best-to-worst ranking, and the lenses are allowed to disagree - that disagreement is the signal. A product that wins under every lens is genuinely dominant. One that tops the value ranking but sinks on reliability settles mid-board, which is exactly where it belongs.

Older boards carry more than four. Two further lenses, performance (outright capability, price no object) and everyday use (what a typical buyer lives with, with no tinkering), were part of the slate until mid-2026. They no longer run on every refresh, but the runs they already played were real and still count. Across the catalogue today a product's record holds anywhere from 2 to 13 runs, depending on how long its category has been here and how often it has been rebuilt.

2. The board includes products you shouldn't buy

Each ranking covers 14 products: the ten that genuinely belong on a shortlist today, plus four traps - real products buyers actually cross-shop but shouldn't buy. The overhyped one, the old favourite coasting on its reputation, the one outclassed at its price. They are ranked honestly alongside the picks, not dumped at the bottom by default, and a trap a run rates above a weak pick keeps that position.

They matter because they keep the tier scale honest: without them, the bottom of the board would just be the worst of a good bunch. With them, D and F mean what they say, and the board does the other half of a tier list's job - telling you what to avoid.

Each run's trap nominations are recorded, so we can count how many independent runs entered a given product as one to avoid rather than as a pick. Where that count is above zero it is shown, on the product page and on the board tile: "2 of 4 research runs" is a different and more useful statement than a low tier, because it tells you the researchers were actively steering you away rather than merely ranking it last.

3. Every placement is cited, and the links are checked

A product only enters a ranking on the strength of what named outlets actually say about it. Each product page lists its sources: who covered it, our reading of the stance they take, and a link to the original coverage. Rankings work from those outlets' current conclusions, not stale ones.

Citing a source is easy to claim and easy to get wrong, so the links are verified by machine before a board is published. Every citation URL is fetched. Links the outlet answers with "no such page" are deleted from the record rather than published, and a product cannot go on a board at all unless at least one of its citations resolves.

Some outlets, RTINGS and Tom's Hardware among them, answer an automated request with an error for every page they have, real or not. We keep their citations and count them as unverified rather than deleting evidence we have no way to test, and the product page says how many of a product's links were actually confirmed. Where a source is cited with no usable link at all, the row stays but says so plainly, and it is left out of the source count.

4. Rankings are scored like ranked matches

Each ranking is a match, and every product in it gains or loses rating points based on where it finished. Place near the top and the rating climbs. Near the bottom and it falls. Mid-field barely moves it. Out-ranking a strong field counts for more than beating a weak one, and a product with little history moves quickly at first, then settles as results accumulate. The runs from one refresh are scored together, against the ratings as they stood before that refresh, so the order the runs happened to finish in cannot change the outcome. No single match decides anything: one lucky first place can't mint a top-tier product. A high tier is earned by finishing high again and again across independent runs. And when a category is re-researched, every match ever played is re-scored from scratch, so the board always reflects the full body of evidence.

5. Tiers come from the rating, not from anyone's gut

A through F are fixed rating bands: reach the rating and the tier is yours, and nothing on a board is ever hidden - a poorly-ranked product just sits low. S is different. It's the apex tier, capped at three slots per category with a high rating floor that only repeated top finishes can clear, so in a strong field the bar for S rises. You don't just hit a number, you out-rank the best. A rating built from very few runs is also discounted toward the middle until further runs corroborate it, which is what stops a single first place from minting an S.

The filtering happens before the board, not on it. A product needs at least two independent runs behind it, and at least one citation link that resolves, before it is published at all: thin evidence keeps a product off the board rather than putting it on with a caveat. Once it is on, every product page shows how many outlets stand behind the placement, so you can judge the weight of the evidence yourself.

6. What is LP?

The LP ("ladder points") number you see next to every product is its position on the category's rating ladder - the same number the tiers and ranks are derived from, just shown directly. Every research run moves it: finish high and it climbs, finish low and it falls, and repeated strong finishes are what push it up. Because rank, tier, and LP all come from one rating, they can never disagree - the #1 product always has the most LP, and a product's LP can only be higher than another's if it also sits above it on the board.

Read it as distance, not just order: two products one rank apart can be 5 LP apart (a coin flip) or 150 LP apart (a gulf), and that gap is information a bare ranking hides. LP is only comparable within a category - 300 LP in office chairs and 300 LP in soundbars say nothing about each other - and it moves whenever a category is re-researched, so expect it to drift between visits as products defend their spots or fail to.

7. The board never freezes

A tier list is only worth anything if it stays current. Categories are re-researched on a rolling basis: each refresh plays the four lens games again over the latest expert coverage and the latest community discussion, so new releases enter the lobby, discontinued products drop off the board, and every rating reflects how a product places today, not how it placed last season. Ranks and tiers shift accordingly - a board you saw last month is not guaranteed to look the same the next time you visit, and that's the point.

Planned next: rating decay. Right now a rating only moves when a product plays a match. Decay will make old results slowly fade between refreshes, so a past champion that stops defending its spot in fresh runs gradually gives way to newer products that are out-earning it. Like any ranked ladder, the system itself gets balance patches: this page describes the process as it runs today, and we'll keep it updated as the model evolves.

8. Check it yourself, by hand or by machine

Everything above is checkable, and not only by reading. Every board is published in machine-readable form alongside the page: append .md or .json to any category URL and you get the same ranking as a table, with the research date, the outlet counts and the tier definitions in the header. Every board in one file and a feed of the most recently researched boards are published too.

We publish these because a ranking nobody can audit is just an opinion with a colour scheme. Quote them, rebuild them, check our arithmetic, or point out a placement the evidence does not support. The terms are on our reuse and citation page: free to use with attribution and a link.

Independence

Rankings are driven only by the published expert coverage, never by affiliate commissions. Some links on this site are affiliate links, but they have no effect on where a product lands. See our affiliate disclosure for details.