Why the loudest master does not win
Every major streaming service normalises playback loudness. If your master is louder than the service target, the service turns it down by the difference — it does not turn quiet content up. So a master pushed to −6 LUFS gets played at the same perceived level as one at −14, except it arrives with all the limiting and none of the dynamics. The loudness war ended the moment normalisation became the default. What you keep from over-limiting is the flatness, not the volume.
What integrated LUFS actually measures
LUFS is loudness units relative to full scale, measured to ITU-R BS.1770-4. The signal is filtered to approximate how human hearing weights frequency, split into overlapping 400-millisecond blocks, and averaged with two gates: an absolute gate that discards anything below −70 LUFS, and a relative gate that discards anything more than 10 units below the ungated mean. That gating is why LUFS tracks how loud a track feels rather than how loud its quietest passages are. This tool implements that algorithm, not an approximation of it.
True peak is not the same as peak
Your DAW meter shows sample peak: the highest value in the file. True peak estimates the highest value in the analogue waveform that gets reconstructed between those samples, which can be higher. It matters because every store re-encodes your file into a lossy format, and that re-encoding can push an already-full signal past full scale. The result is distortion that was never in your master. This tool oversamples four times to measure it, the same way a broadcast meter does.
The number to aim for
Master to around −14 LUFS integrated with true peak at or below −1.0 dBTP, and you will sound the way you intended on every major service. If you deliberately go louder than −14 LUFS, take true peak down to −2.0 dBTP to leave room for the encoder. If your peak-to-loudness figure is under 8 units, you are limiting hard enough that normalisation will erase the benefit. None of this is a rule you must follow — it is simply what the platforms do to your file after you upload it.
The other things this checks
Sample rate and bit depth against the accepted set. Channel count, because mono files and files where both channels are identical are common rejection reasons. Clipping, counted in samples. DC offset. Leading and trailing silence, because more than five seconds at either end is rejected by several services. And stereo correlation, which catches phase problems that make a mix collapse when it is played back in mono — which is most phone speakers, most smart speakers and most club systems.
Which file types work?
Anything your browser can decode: WAV, FLAC, MP3, AAC or M4A, and AIFF in most browsers. For a real delivery check use the actual WAV or FLAC file you plan to submit, because analysing an MP3 bounce tells you about the MP3, not about your master.