Sample data. Every keyboard on this site is illustrative — nobody has typed on these boards, the measurements are constructed and the photographs are generated placeholders. The whole site is set to noindex until real reviews replace them. What this means
Bottomout.

Budget mechanical keyboards, actually tested.

Every board here was bought at retail, used for at least two weeks, and measured the same way — sound at a fixed mic distance, latency against a photodiode, actuation force on a gauge. The bad ones stay published.

10
boards tested
998
days of use logged
33
head-to-head comparisons
0
review units accepted
75%Tactilehot-swap

Ferrule 75

Highest overall score here at 8.5. The one I recommend without caveats.

8.5/10$89
TKL (87-key)Linearhot-swap

Tessera TKL

Quietest board measured, 51.8 dB at 30 cm.

7.5/10$84
60%Clicky

Drift 64

Cheapest board tested at $22 — and the review says exactly what that buys.

4.8/10$22
75%Tactilehot-swap

Palisade 75

Lowest score on the site at 4.4. Left up on purpose.

4.4/10$64

Depends what you need it for

All lists →

How this site works

Bought, not sent

Every board was paid for at retail. No review units, no loans, no early samples — which means no relationship to protect when a board turns out to be bad.

Measured the same way

Same room, same mic distance, same desk mat, same 500-sample latency run. The absolute numbers are less useful than the fact that they are comparable to each other.

Scored by formula

Six criteria with published weights. There is no override field anywhere in the data, so no board can be moved up a list. See the formula.

Failures stay up

One board here was returned after nine days. Its review is still published, because a review site that only lists boards worth buying is a catalogue.

Which keyboard specs actually matter

Product listings are full of numbers, and most of them are noise. Which specifications predict whether you will like a keyboard, which are marketing, and which cannot be trusted at all.

How to make a cheap keyboard sound better

Five modifications ordered by how much they actually change the sound per hour spent, measured on the boards tested here — and the two that are usually a waste of time.