Methodology
How I test
Every review links here, because a score is meaningless without the method behind it. This page is the method.
This site is running on sample data
All 10 boards currently listed are illustrative. Nobody has typed on them, the measurements were constructed to exercise the system, and the photographs are generated placeholders that say so inside the image. The whole site is served noindex while this is true.
Everything else on this page describes the method that real reviews will follow — and the build refuses to treat the site as live until every sample entry has been replaced.
The short version
I buy the board at retail, use it as my only keyboard for at least two weeks, measure four things the same way I measure them on every board, and score six criteria with published weights. Then I write down what went wrong, because that is the part that is hard to find anywhere else.
Acquisition
Every board is bought with my own money at the price anyone else would pay. I do not accept review units, loans, or pre-release samples. This is not a moral position so much as a practical one: a free board creates a relationship, and a relationship is exactly what makes a bad review awkward to write.
If that ever changes — if a brand sends something unsolicited and I decide to test it — the review will say so at the top, in the same place, every time.
What gets measured
Four figures, taken identically on every board. The absolute values matter less than the fact that they are comparable to each other, which is why the setup never changes.
| Sound | Peak dB(A), microphone at 30 cm and 45°, averaged over 200 keystrokes of ordinary typing. Same room, same desk mat, same time of day. Room noise floor is measured before each session and a board is re-tested if the floor moves more than 2 dB. |
|---|---|
| Latency | Click-to-photon, median of 500 samples, measured with a photodiode against a high-frame-rate capture. Taken wired, and again over each wireless mode the board supports. Reported as the median rather than the best figure, because the best figure is not what you type on. |
| Actuation force | Grams at the actuation point, measured on the E key with a force gauge, three runs averaged. Compared against the manufacturer's rating — which is frequently optimistic. |
| Case flex | Millimetres of deflection at the centre of the board under a 5 kg load, measured with a dial indicator. This is what separates a genuine gasket mount from a marketing claim. |
The scoring formula
Six criteria, each scored 0–10, combined with fixed weights that sum to 1.00. Every review shows the full breakdown, not just the total.
| Build — 20% | Case rigidity, plate flex, screw quality, and how it survives being picked up one-handed. |
|---|---|
| Typing feel — 25% | Consistency across the board, wobble on the big keys, and how the mount behaves under a hard bottom-out. |
| Sound — 20% | Pitch, hollowness, and rattle. Measured at a fixed mic distance and also judged by ear. |
| Stabilisers — 15% | Ticking, rattle and mushiness on space, shift and enter — the most common budget failure. |
| Software — 10% | Remapping without an account, without a cloud service, and without a driver that only runs on Windows. |
| Value — 10% | What it costs against what else that money buys today. Re-scored when prices move. |
The ranked lists at /best re-weight those same criteria for a particular use — a typing list weights typing feel and stabilisers, a gaming list adds a bonus computed from measured latency and polling rate. Each list states its own weights on the page.
There is no override
The data model has no field for manually placing a board on a list. Ordering is computed at build time from the scores, so moving a product up would mean changing its score — which would change it everywhere at once, including on its own review. That is deliberate: it makes quiet favouritism impossible rather than merely against policy.
Scoring is relative to price, once
Five of the six criteria are absolute — a rattling stabiliser scores badly whether the board cost $22 or $95. Only Value considers price, and it is re-scored when prices move rather than left at whatever was true on launch day. This is why a $22 board and a $95 board can both score reasonably: they are being asked different questions in exactly one column.
Prices
Every price on this site carries the date it was checked. Past 45 days it is flagged as stale on the page rather than left looking current. Nothing here scrapes live prices, because a scraped price that breaks silently is worse than a dated one that is honest about its age.
Long-term updates
Boards get revisited. Each review carries a dated timeline of what changed — stabilisers that got worse, keycaps that shined, sockets that loosened, gaskets that compressed. Entries are appended, never edited away, and a score that changes says when and why. Several of the most useful notes on this site are of the form "nothing changed", which is itself worth knowing at month five.
Who is writing this
A. Editor — Reviewer, Bottomout. Buys keyboards, measures them, and writes down what went wrong.
- Every board on this site was bought at retail with my own money. No review units, no loans, no gifted samples.
- Boards are tested for a minimum of two weeks of daily use before a review is published.
- Sound is recorded in the same room, at the same mic distance, on the same desk mat, for every board.
What I am not
Stating the limits is part of the method, not a disclaimer bolted to the end of it.
- I am not an acoustics engineer. Sound figures are comparative between boards on this site, not absolute laboratory measurements.
- I do not test long-term switch durability. Rated lifespans come from the manufacturer and are not verified here.
- I do not disassemble boards under warranty, so internal build notes are limited to what is visible without breaking a seal.
How the money works
Some links on this site are affiliate links. The score is computed before any link is attached, the lowest-scoring board on the site is still published with its review intact, and one board here was returned after nine days and kept up. Full disclosure.
Corrections
If a measurement here is wrong, tell me and I will re-run it — and the timeline on that review will record both the original figure and the correction. Contact.
This page describes Bottomout's method as of August 2026. Changes to the criteria or their weights are noted here with a date.