Loudness targets are a floor, not a ceiling

The advice goes: streaming platforms turn loud masters down, therefore master quieter. It is half right, and the half that is wrong produces worse records than ignoring the whole thing would have.

What normalisation actually does

Playback normalisation applies gain. That is the entire mechanism. The platform measures your track's integrated loudness, compares it to their target, and applies a fixed gain offset so it plays back at a consistent level alongside everything else.

It does not undo limiting. It does not restore transients. It does not recover dynamics that were removed before upload. A master crushed to minus six LUFS and turned down eight decibels does not become a master made at minus fourteen — it becomes a crushed master, playing quietly.

Normalisation punishes over-limiting. It does not punish loudness.

That distinction is the whole thing, and it gets collapsed constantly. Loudness is a level. Over-limiting is a process. Normalisation removes the reward for the first while leaving all the damage from the second in place.

The competition that no longer exists

For roughly two decades, there was a real and rational reason to master loud. Records played back at whatever level they were mastered at, in a playlist, on the radio, in a shop. A quieter record sounded worse next to a louder one — not because quiet is worse, but because human hearing reliably mistakes louder for better in a direct comparison.

So mastering engineers were asked to win that comparison, and the tools got very good at it, and a lot of records from that period are exhausting to listen to all the way through.

Playback normalisation ends that competition. Your record and the loudest record ever made now play back at the same measured loudness on the platforms most people use. There is nothing to win. Limiting for competitive level in that environment is paying a real cost for a benefit that has been switched off.

So why not just master to the target

Because the target is not a specification. It is where the platform's gain calculation lands, and it varies — around minus fourteen LUFS on several services, minus sixteen on Apple Music, different again for broadcast. Some listeners have normalisation turned off. Some download. Some contexts, notably club playback and CD, have no normalisation at all.

More importantly, some records should be loud. A dense, aggressive, deliberately saturated record is loud because that is what it is. Making it quieter to hit a number does not make it more dynamic — it makes it the same record with less of the thing that defines it.

The target tells you what the platform will do. It does not tell you what the record should be.

A better question than how loud

Instead of asking what number to hit, ask what the record can survive.

Turn the limiter down until it is barely doing anything and listen. Then push it and listen again. Somewhere there is a point where you gain perceived energy and lose something you care about — the snare stops cracking, the vocal stops breathing, the low end stops moving. That point is different for every record and it is the only number that matters.

The right loudness is the loudest the record can be without losing the thing that makes it good.

If that lands at minus nine, the record is a minus nine record and the platform will turn it down. If it lands at minus sixteen, it is a minus sixteen record and the platform will turn it up. Both are fine. Both play back at the same level for most listeners.

True peak, which is a real constraint

Loudness targets are advisory. True peak is not, and it is where the genuine technical requirement lives.

A digital file can contain peaks between its samples that exceed the highest sample value. Reconstruct the waveform — which every converter and every lossy decoder does — and those inter-sample peaks appear. If your file peaks at zero dBFS on the sample meter, the reconstructed signal may well exceed zero, and the decoder will clip it.

Lossy encoding makes this worse, because the encoder is approximating and its approximation overshoots. A file that is clean at zero on a sample meter can come out of an encoder audibly distorted.

This is why minus one dBTP is the common recommendation. It is not superstition and it is not headroom for its own sake — it is room for the reconstruction and the encoder to do what they do without clipping.

Which means, when the loudness target and the peak ceiling disagree, the peak ceiling wins. You can be quieter than the target with no consequence at all. You cannot exceed the ceiling without audible damage on some playback systems and not others, which is the worst kind of problem because you will not hear it on yours.

What to actually deliver

Deliver one master, made at the level the record wants to be, with true peak under minus one, at twenty-four bit and the session's sample rate. Let each platform apply its own gain.

Platform-specific masters are almost always unnecessary. The exceptions are real but narrow: a club or DJ master that needs to be loud and will not be normalised, a CD master with its own constraints, a broadcast deliverable with a mandated target and a hard specification.

Do not master at twenty-four bit and then dither to sixteen before uploading. The platform will transcode from whatever you give it, and giving it the highest-quality source means its encoder has the most to work with.

The short version

Stop limiting for competitive level, because the competition ended. Keep the record as loud as it wants to be, because normalisation does not punish that. Respect the true peak ceiling, because that one is real. Deliver one high-resolution master and let the platforms do the arithmetic.

The number on the meter was never the point. It was a proxy for winning a comparison that no longer takes place.