Is this just a preset with a different name on it?
No, and the dataset on this page shows the difference. The correction is computed per track: your spectrum is measured in 31 third-octave bands, compared against the target, and the difference applied band by band within a bound of +6 and -9 dB. Two measurements make that concrete. The "no EQ" profile lands 0.03 dB from its target, meaning it genuinely does nothing to your tone, while the shaping profiles move the same source from 3.83 to 5.38 dB away down to 1.71 to 2.25 dB away. And the true-peak ceiling is not a setting at all: it is found by encoding your file and measuring what comes back. On the demo above it stayed at -1.0 dBTP at three levels and had to move to -1.76 dBTP at another.
What happens to my music after I upload it?
It is written to disk on the machine running the job, processed, handed back to you, and deleted 7 days later by a sweeper that runs every hour. Nothing is used to train anything, because there is no model here to train: the whole engine is signal processing on a CPU. We claim no rights to your audio. The part we will not dress up: there is no delete-now control yet, so 7 days is the shortest answer available today. The privacy page lists that and the other gaps rather than rounding them off.
Should I upload the WAV export or the share MP3?
The WAV export, every time. What the share encoder threw away is gone and no mastering brings it back, ours included. The analyzer measures where your spectrum stops and prints that number: on the file behind the demo above it found 20.1 kHz and graded it as fine. A much lower wall is the signature of a share file, and the report says so instead of quietly boosting bands that contain only noise. We have not published a measurement of what a typical share file walls at, so the tool gives you the figure for your own file rather than a rule of thumb.
Why is my master quieter than the level I asked for?
Because we print what the limiter achieved, not what you requested. Ask the demo above for -6 LUFS on that file and what comes out is -9.31 LUFS after 18.68 dB of gain reduction: the material runs out of room, and the only thing left to squeeze is the music. At -16, -14 and -12 LUFS the same file lands exactly on target with 0.0 dB of reduction. A service that always reports the number you typed is reporting your input back to you.
Why does my master not sound louder than the other service's at the same LUFS?
Because at the same integrated loudness it is not louder. LUFS is the loudness, and every normalising platform turns both files to its own target anyway: in the dataset above Spotify lands the master at -14.0 to -14.0 LUFS no matter which of the six levels was delivered. What differs between two masters at the same LUFS is dynamics and tone, not volume. We have not measured any competitor's output, so we do not claim to be louder than one, and anybody who does claim it either measured it or made it up.
Can I use my own reference track instead of a profile?
Yes, on the paid tier. A reference is analysed exactly like a style profile, so your track becomes the target curve, and the same +6 / -9 dB per-band bound applies. Two things it will not do. It does not clone a mix: matching a spectrum is not matching arrangement, and if your reference is far away the correction stops at the bound rather than chasing it. And a reference that is itself lossy hands you its own ceiling, which the report will point out.
Does album mode do anything a batch button would not?
Yes, and it is measurable. The target is the release's own median spectrum, so coherence costs each track the smallest possible correction rather than dragging all of them toward a preset. On the 3 tracks measured further up this page the tonal spread went from 1.50 to 1.09 dB per band. The loudness spread going from 1.02 to 0.00 LU is not the achievement: any per-track normalisation does that. The tonal figure is the one that separates album mode from running the same job three times.
Can you guarantee no platform will distort my track?
We guarantee what we measured, which is not the same sentence. Every platform in the report gets its own encode at its own bitrate and its true peak measured on the decoded result, and the ceiling is lowered until that measurement is inside the limit. What we cannot cover: platforms change encoders without announcing it, some of these targets are industry consensus rather than published figures, and a platform that re-encodes a second time downstream is outside anything we can see. The platform table on this site marks which numbers are published by the platform and which are not, and it is honest about how few are.