Peak level
Also known as: true peak, maximum level
The highest instantaneous value a track reaches, which is what decides whether it distorts rather than how loud it sounds.
Updated
What it means
Peak level is a single worst case number. It says how close the loudest instant comes to the ceiling, and therefore whether the recording is at risk of distorting rather than how loud it will seem.
It is routinely confused with loudness and controls something entirely different. A quiet track with one sharp transient has a high peak, while a dense loud track can have a lower one.
Both matter and they are managed separately. Peaks are limited to prevent Clipping, and perceived level is set afterwards against a LUFS target, because neither measurement substitutes for the other.
How it works in practice
During recording the peak is what headroom protects. Setting levels so the loudest expected sound lands several decibels below the ceiling survives an unexpected Impact hit or gust.
In the mix a limiter catches whatever remains. Shaving a handful of transients is inaudible, while leaning on a limiter to raise the whole track is compression by another name and audibly flattens the result.
True peak measurement is the version worth trusting, because a signal can exceed the ceiling between samples and arrive distorted after conversion even when the sample values looked safe.
What to watch out for
Mastering to exactly the ceiling leaves nothing for the encoder. Lossy compression shifts values slightly, so a track at zero can clip after platform processing even though the file was clean.
Watching peaks alone also produces timid recordings. Levels set so nothing ever approaches the ceiling put the useful signal far too low, and raising it later raises the noise with it.
A peak also says nothing about what caused it. Wind, a bump against the housing and a genuine landing look the same on a meter, which is why Audio peak data needs picture to interpret.
How RawClip handles it
The finished mix is limited so nothing reaches the ceiling, with a margin left for what platform encoding does to the file after upload. Perceived loudness is set separately against its own target, because preventing distortion and matching level are two different jobs with two different numbers.