How we test audio gear

The rules every review follows

• Time on the chain. At least two weeks of daily listening before a score is committed, typically four to eight weeks for flagship gear, and never a first-impressions verdict.
• Against a reference, at its price. Every piece is compared head-to-head with the reference unit for its category and price tier, on the same chain, switched in and out instead of held against a memory of one.
• Sighted, with levels set by ear. The writer knows which product is playing and sets the volume by ear, and no review reports a blind test. What that means for a small difference is set out below.
• Figures from the maker. No review publishes a measurement made here. The specifications a review quotes are the maker's published figures, and the review names the standard that defines them.
• One published rubric. A score is the writer's judgment on the same weighted axes in every category, and what earns the points on each axis in a category is printed on that category's hub. How a score is built and what each verdict band means.

The listening room

Jakub Charkiewicz's listening room, where the reference chain lives, has been treated so that what a review describes is the gear rather than the room. It matters most for the categories heard through a room (speakers, DACs, speaker amps & preamps, digital sources and accessories), and each guide in those categories says so in its own testing notes.

• Front wall. Ekustik Woody Queen Performance Bass panels directly behind the speakers - 15 cm deep, with a dense recycled-PET core the maker rates at NRC 0.94 (ASTM C423) and effective from 100 Hz - so bass reflecting off the front wall no longer collides with the direct sound. Source: Ekustik Woody Queen review.
• First reflections. Ortvik Parametric Towers standing free at the first reflection points (EBU Tech 3276), where they absorb and diffuse the early reflection before it reaches the listening seat and smears the direct sound. Source: Ortvik Parametric Tower review.
• Decay. Between them the panels and towers bring the room's reverb time down without leaving it dead, so micro-detail the speakers produce is heard rather than masked by the room.

Power and digital

The same system feeds every review heard on the reference chain, headphones included, so the noise floor and the digital front end are the same for every product compared on it.

• Mains. A Synergistic Research PowerCell 14 conditions the mains for the whole system - the noise floor every review sits on. Source: Synergistic Research PowerCell 14 review.
• Power cables. Every power cable in the system replaced with Monosaudio P902E6R, and the USB cable upgraded with them. Source: Monosaudio P902E6R + USB Cables review.
• Grounding. An AudioMica Terra Consequence grounding cable, for a darker background with nothing playing. Source: AudioMica 4-all-in-1 Cable Set review.
• Transport. A Denafrips Hermes 12th digital-to-digital converter reclocks the computer's USB stream against its own oven-controlled clock before any DAC sees it. Source: Denafrips Hermes 12th Anniversary review.
• I²S link. DH Labs cable on the I²S connection between the transport and the DAC. Source: DH Labs, iFi, Vera-Fi HiFi Tweaks Roundup.
• Library. A lossless PCM music library, played bit-perfect; MQA is not used.

The reference chain, category by category

What each category is heard through. Every component is a product this site has reviewed, linked to its own verdict, and each hub prints the same chain above its reviews.

How a listening session runs

The same recordings, across multiple genres and multiple sources, return in every session, so a product is heard on material the reviewer already knows in detail. The reference is swapped in and out on the same chain with levels set by ear, and the listening is spread over weeks of daily use, because a flaw that shows up on one track in one mood is not the same as a character a listener lives with. The buying guides record what their own sessions looked for (the sources an in-ear was heard on, the loads an amplifier drove, the recordings chosen to expose a speaker) in each guide's "How we tested" block.

The listening is sighted: the writer knows which product is playing, and sets the volume by ear for each one. Knowing the brand, the price and the look of a product changes what listeners report, which is what Harman's trials found when the same loudspeakers were rated sighted and then blind (Sound Reproduction (Toole, Olive and Welti)), and a level difference too small to notice as loudness can be heard as better sound. So the large differences a review describes, a headphone's tonal balance or a speaker's bass, rest on firmer ground than the small ones. What a review reports for a cable, a power product, a network switch, a digital transport, or between two competent DACs, is the writer's impression: no review here reports a blind check, which is the standard way to confirm a small difference (ITU-R BS.1116-3). The site's ABX test runs one at home, with the two files level-matched.

Where the figures come from

No review publishes a measurement made here. The impedance, sensitivity, output power, noise and distortion figures a review quotes are the maker's published specifications, checked against the maker's own page, and a figure found wrong is fixed on the review and listed in its Corrections box. Each review's specs card names the standard that defines its figures: a converter's, a digital source's or an amplifier's noise and distortion figures follow the definitions of AES17, and each other category's the standard its card names. A spec sheet is an input to a review and never its verdict; where a review holds a figure against the listening, the listening decides the score.

Recording gear

A microphone is recorded, not played back, so it is judged on a real voice in a real room (where its response is lifted or dipped, how much hiss it adds, how much of the room comes with it), and where a comparison is the point it is recorded side by side. The FIFINE K688 review reads the same sentences three ways, over USB, over XLR and against a $1,300 Neumann TLM 103, so the difference is on the recording rather than in a claim. The reviewers record on their own chains, which is why the microphones hub names no single one.

How far to trust a score

What a score is

A score is one writer's overall judgment of a product after weeks of listening, set against the reference for its category at a similar price. The five weighted axes of the rubric say what that judgment weighs; they guide the number, the number is not computed from them, and no review records a separate score per axis. The breakdown on a review's At-a-glance card is derived from the review's own pros and cons and adds back up to the published score: it shows where the review places a product's strengths, and it is no second opinion. The weights have not changed since they were first published, on May 12, 2026.

A score is also one listener's preference. A headphone one writer finds detailed and exciting can sound harsh to a reader more sensitive to treble, so where a review gives the material for it, its FAQ says who the verdict does not suit.

How closely two writers agree

Each product here has been scored by one writer. The Audio Stuff has two writers and no product has been scored by more than one of them, so there is no figure yet for how closely two writers agree. If a product is ever scored by two, both scores are published and this page reports how often they fall within half a point.

What a gap between two scores means

A score has no measured margin of error. The scale moves in tenths so that closely matched products can be ranked, and a comparison calls a one-tenth gap a near-tie. A wider gap within a category, 7.8 against 8.4, is the writer's judgment that one product is clearly better at what the rubric weighs, and the reasons are in the two reviews' pros, cons and comparisons. The same gap says less between two writers, and less again across categories.

Scores compare within a category, each set against its own category's reference. Nothing tests whether an 8.5 headphone and an 8.5 DAC are equally good, so across categories only the verdict carries one meaning, and the verdict bands are this site's own. A Reference verdict means a product sets the standard in its class, in any category.

When a score changes

A published score is not recomputed when a price moves or a newer product arrives. When the editor changes one, the review keeps the earlier score under Score history, with the date and the reason. A tracked store price never changes a score. So far 1 of the 71 reviews has had its score changed since publication: the HIFIMAN HE-R9 (7.4 to 6.9 on Jun 15, 2026). Every score and its history is also in the open data, ratings.json.

Who checks the scores

No outside lab, auditor or second publication checks these tests or these scores. What a reader can check is on the page: the video reviews and the products each review was compared with (below), the chain each category was heard on (above), the published figures and their standards, and the corrections and score changes each review lists.

The evidence on every review

Every one of the 71 reviews has a video review on the channel, where the product is shown and heard, and 68 of them link written sections to the moments in the video that cover them (245 links in all). The photographs in a review are taken from its own video review, of the unit that was reviewed, except an image whose caption credits its maker. Every review names the products it was compared with, so a reader who owns one of them can reproduce the comparison.

Jakub Charkiewicz has written 66 reviews and Bastien Beauchamp has written 5 reviews; figures as of Oct 28, 2026, the day the latest review was published.

Where review units come from

Some units are lent by their manufacturer or distributor and go back when the loan ends; others are bought. Where a review records it, the unit's source is stated in the review's header. A loan never commits the site to a verdict, pre-publication review is never granted, and the site carries no advertising and no affiliate links, so a negative review ships with nothing for a brand to withdraw.

Corrections

A substantive correction lands on the review it corrects, dated, and moves the review's Updated date; a check that finds everything still accurate records a Checked date instead (what each date means). 23 reviews carry a dated correction, the most recent from Sep 27, 2026. Anyone can report an error.

Standards

Every standard the reviews cite, with what each one is used for, is listed under Standards we cite.