How we work
Two things must be reliable: the file you export and the figures we publish. Here is how we check both, and the limits of those checks.
Updated October 16, 2026
What runs in your browser
The editor's audio engine is compiled to WebAssembly and runs on your device: trimming, fades, tempo detection, stretching, loudness measurement and encoding.
Rendering happens join by join in temporary browser storage, followed by a single loudness measurement and a single final encode. Temporary files are deleted when the export finishes or is cancelled.
Loudness normalisation
When normalisation is on, the whole mix is measured to the EBU R128 standard, then one gain brings it to -14 LUFS integrated, with peaks kept under -1 dB.
That gain applies to the whole file: it sets the overall level of the mix, it does not fix the volume difference between two songs. If an older track sounds quieter than a recent one, pick a remastered version or listen to each join before exporting.
How we test the engine
The tests run automatically against the real audio engine, in a real browser, on every code change.
- 22 render cases: formats, fades, hard cuts, digital silence, tempo sync from 90 to 120 BPM, manual stretching.
- Computer-generated test files (sine waves, click tracks at 90 and 120 BPM): the expected result is known to the sample.
- Every export is decoded independently, then checked: duration, signal present at the start, middle and end, fades, loudness at -14 LUFS within 1.5 LU, no clipping.
- Beat sync: after a 90 to 120 BPM transition, the click positions must not vary by more than 50 ms (standard deviation).
- 14 scenarios in the real interface (import, order, trimming, cancelling, twelve tracks, a six-minute mix, retry after a failure) in Chromium, Firefox and WebKit, plus a mobile viewport.
- Memory tests: 11 three-minute songs exported under a 96 MB WebAssembly memory limit, to simulate a modest device.
What these tests do not cover
The mobile viewport is simulated: a real iPhone on Safari and Stripe's hosted payment page are checked separately, by hand. The test files are synthetic: they prove the arithmetic is right, not that a transition sounds good on two particular songs. For that, nothing replaces listening to your whole mix.
Where the published figures come from
Our guides and use-case pages quote a lot of figures: competition lengths, tempo, levels. Each one falls into one of three categories, and we present it as such.
- Rules: required lengths and music rules, taken from the official rulebook or, failing that, from a press source that reports it. The link sits under the table.
- Measurements: technical values (loudness, tempo, fade length) checked with the engine itself.
- Guidelines: number of songs, excerpt length, sample structure. These are editing recommendations, not rules, and we say so.
- Every page shows an update date. Rules change every season: the official version always prevails.
Report a mistake
A wrong figure, an outdated source, a rule that changed: write to us. The correction is published with a new update date.