Loudness normalization
The rule that decides how loud a record actually sounds to a listener, and it is no longer set by whoever mastered it. Both halves of the chain are now published documents: ebu-r128 for broadcast (−23 LUFS, true peak ≤ −1 dBTP) and each streaming platform's own target, of which spotify-loudness-normalization (−14 LUFS) is the one this wiki holds.
The measurement came first
LUFS — Loudness Units referenced to Full Scale — is defined by ITU-R BS.1770, along with the level-gated integration that measures foreground loudness rather than averaging in the silence. Everything else is policy on top of one meter. That ordering matters for the argument this spoke keeps having: before BS.1770 there was no agreed way to say one master was louder than another, so the loudness war was fought with peak meters, which by the EBU's own account "do not reflect the loudness of an audio signal."
What it does to a mix
A platform that normalises turns a hot master down. Spotify says so directly: a master above the target gets negative gain, and no distortion is added — the playback is simply the master, quieter. So the limiting and clipping used to win loudness are still in the file, still audible, and the loudness they bought is removed on the way out. Whatever case there was for crushing a master into a brick, this removes the part where it made the track stand out.
The corollary is the one worth putting in front of anyone using the plugins this wiki pages: the dynamics processing on a master bus is now a sound-design choice rather than a competitive one. OTT-style multiband compression (see audio-effect-plugin) does something to a mix that a listener can hear; making the mix louder is not part of it any more.
Two targets, one meter
−23 LUFS and −14 LUFS are nine units apart and neither is wrong. Broadcast programmes run long, sit inside a schedule with adverts and trailers, and are watched in rooms with a noise floor; music streaming is per-track, and the platforms landed near the level a typical modern master already had. The interesting thing for this spoke is that both numbers are published, checkable, and enforced by software after the mastering engineer's last decision — the point at which "how loud should this be" stopped being an aesthetic question and became a delivery spec.
Related
analog-vs-digital — the spoke's founding argument, which this touches from an unusual angle: here the digital side ended a decades-long degradation of recorded sound rather than causing one.