Every streaming platform turns your record down to about the same level on playback. Spotify says -14 LUFS. Apple says -16. YouTube lands close to -14. So for about eight years the advice going around has been: stop fighting, master to -14, the loudness war is over.

I wanted to know whether anyone actually did that. We have measurements on 144,645 tracks with a release date between 2014 and 2025, the same analysis that runs under the Sonic Analyzer and the same table the tonal balance by genre study came from. Integrated loudness, loudness range and true peak on every one of them, split by release year, by how big the release got, and by genre.

Nobody masters to -14. Not in 2016, not in 2025, not the big records and not the bedroom ones.

The number has not moved in twelve years

Medians, so a few crushed records cannot drag a year. Read the middle column top to bottom. The loudest year in the set is 2016 at -11.6 and the quietest is 2021 at -12.3. That is 0.6 dB across twelve years, less than the difference between two limiter presets. The whole industry was told to come down two decibels and it came down half of one, then stopped.

The share of records below -14, the ones that actually took the advice, went from a fifth to a quarter. The share louder than -9, the ones that never heard it, is 6 to 8 percent every year without exception. The war did not end. It froze.

The true peak column is the one I did not expect. Half of all releases peak at or above 0 dBTP in every year of the set. The standard advice is a ceiling at -1 dBTP so the encoder has room, and half the records on the platforms ignore it. If you have ever wondered why a song sounds a little harsher on your phone than it did in the studio, that column is a reasonable place to start looking.

Big records are louder than small ones, by less than you think

I expected the hits to be the loud ones and the home masters to be the timid ones. Split the same years by how big the release got:

The big records are louder, by about half a decibel, and in 2025 the gap closed completely. Loudness is not what separates the records that broke from the ones that did not. It has been a solved problem at every budget for a decade. What separates them lives in the balance, which is a different post.

By genre, where your record is expected to sit

This is the table I would print out. Median integrated loudness, the loudness range, and how wide the stereo image measures, for the genres with the most releases in the set.

Three things in that table are worth more than the whole loudness argument.

Hip-hop is quieter than metal. The genre everyone assumes is the loudest sits a full decibel under metal and emo, below pop, below country. And it is the narrowest music on the list: a median width of 0.19 and a left-right correlation of 0.93, which is most of the way to mono. The weight of a rap record comes from the center, from the kick and the 808 and the voice sitting on top of each other, not from spreading things out. If you are mixing hip-hop and reaching for a widener, the records you are competing with did not.

Metal is the loudest and the widest of the band genres. Those two things go together in a specific way: the guitars live hard left and hard right, the center carries kick, snare, bass and voice, and the master can go to -11 without the center turning to mush because the sides are already out of its way.

Ambient is the widest thing in the set and the quietest by seven decibels. Nobody is competing for loudness there, so the whole dynamic range comes back: a loudness range of 4.5 LU against 1.5 for metal. The quieter the genre, the more room it has left.

The honest version of "how loud should my master be" is: look up your lane in that table, and then look at the "below -14" column. If you are making jazz and you master at -11, you are the loudest record on the playlist and it will sound wrong, not impressive. If you are making house and you master at -14, you are quieter than 86 percent of your neighbours and the normalization will not save you, because it only turns records down.

What I take from it in my own room

I do not master to a number. The tonal balance gets right first, on the small speakers and in mono, and the level is the last decision, not the first. The meter confirms what the record can take; it does not get a vote on what the record should sound like.

What the data changed for me is the peak. I used to treat a -1 dBTP ceiling as a courtesy. Half the industry does not bother, and I now think that is where a lot of the "it sounded fine in the session" complaints come from. The ceiling stays.

And it took the pressure off the number itself. The hits are not louder than everyone else. They have not been for a decade. If a client asks me for -9, I show them the 7 percent row and ask what they are hoping it buys.

Method

Every track was analyzed from audio by the same pipeline that powers the Sonic Analyzer, and these numbers are in the table the analyzer benchmarks your scan against. Loudness is integrated LUFS (ITU-R BS.1770), loudness range is LRA in LU, true peak is dBTP. Stereo width is the ratio of side to mid energy; correlation is the left-right correlation coefficient, where 1.0 is mono. Release dates come from the catalogue metadata and a track enters the year of its release. Size groups use the track's Spotify play count at the time of measurement. All figures are medians or quartile counts, never means. Tracks measuring outside -30 to -3 LUFS were excluded as bad captures. The by-year table covers 144,645 tracks; the genre table uses the full analyzed set and shows genres with at least 2,000 tracks.

Want to know where your own master sits against your lane? Run it through the analyzer. It reads loudness, true peak and width in about a minute and tells you how far you are from the records around you.