# Loudness Desk > MASTERING LOUDER DOES NOT MAKE YOU LOUDER. Every platform measures your integrated loudness and > applies one gain to reach its own target - `gain = target - measured` - which undoes exactly and > completely whatever was done to make the file loud. A master at -6 LUFS is turned down 8 dB to > reach Spotify's -14 and plays back at EXACTLY the same loudness as a master at -14 that was > turned down by nothing. The two differ in one respect: the loud one has 8 dB less room to move. > Five lanes over one set of measurements, plus a free browser-side engine in exact decibel > arithmetic against ten real delivery targets. URL: https://loudness-desk.skillsafe.ai/ API: https://loudness-desk.skillsafe.ai/api.html Derived from affaan-m/everything-claude-code (https://github.com/affaan-m/everything-claude-code), MIT: its `video-editing` skill covers AI-assisted editing workflows from raw capture through FFmpeg and Remotion to final polish. Not affiliated with or endorsed by affaan-m. ## The one line everything follows from ```text gain the platform applies = its target - your integrated loudness ``` One subtraction. It is the whole mechanism, and it settles the loudness war: two masters of the same music, one at -6 and one at -14, play back at IDENTICAL loudness on every normalising destination. The loud one paid 8 dB of dynamic range for the tie. NEVER SAY A MASTER IS "TOO QUIET FOR STREAMING". On a platform that only attenuates, a quiet delivery has nothing done to it at all - it plays quieter than its neighbours, which is a choice about how the record sits and not a fault. The destinations that DO raise quiet material are the ones to watch, and for the opposite reason: they raise the peaks too. ## Peaks SAMPLE PEAK IS NOT TRUE PEAK. A sample meter reads only the values in the file; the waveform reconstructed on the way to a speaker passes BETWEEN those values and goes higher, routinely by a dB or more. A file reading exactly 0.0 is already over, and "the meter never hit zero" is not evidence about true peak, because the meter that said so was not measuring it. A LOSSY ENCODER DOES NOT PRESERVE PEAKS. It approximates the waveform and the approximation overshoots. That is why every streaming spec asks for a ceiling below zero: the headroom is not politeness, it is the room the encoder needs. NORMALISATION GAIN LANDS ON THE PEAKS TOO. Decibels are logarithmic, so the same subtraction applies to both - a platform that raises a quiet delivery raises its peaks by the same amount, and a file delivered legal can be clipped by the platform being helpful. That failure is invisible in the studio: every meter in the room reads correct, the delivered file IS correct, and the distortion is created afterwards by a process nobody in the room controls. A peak already over the ceiling and a peak pushed over by normalisation read the same on a meter and have different fixes. Say which one you found. ## Track mode and album mode are different records Under TRACK normalisation each song is measured and gained separately, so all of them arrive at the target and the differences between them are flattened. Under ALBUM normalisation one gain is applied to the whole set, so the differences are preserved exactly as delivered. The same numbers are therefore right under one mode and wrong under the other. Three songs at identical loudness is correct for a singles-driven streaming release and is not a record anybody mixed - the ballad was supposed to be quieter than the single. ASK WHICH MODE before quoting a single number. ## Broadcast is a spec, not a preference Streaming targets describe what a platform will DO to your file. R128 and A/85 describe what will be REJECTED: deliver outside tolerance and it comes back rather than being quietly normalised. So one master cannot serve broadcast and streaming. -23 LUFS is nine decibels below a streaming target: a file that satisfies the spec is far quieter than streaming expects, and one mastered for streaming fails the spec outright. When both are on the sheet the answer is two masters, and the only question is which is the source. Broadcast and Netflix targets are dialogue-gated in practice - the measurement follows the speech rather than the programme, which is why a film can pass on programme loudness with dialogue nobody can hear. The figure is right and the thing is wrong. ## What this page cannot do Everything here is arithmetic over measurements you supply. THIS PAGE CANNOT HEAR ANYTHING. It has no opinion about a mix, does not know what the material is, and cannot tell a squashed record from a loud one that was always meant to be loud. So never describe how something sounds: `-9.8 dB` is the attenuation, not "crushed"; `2.1 LU` is the loudness range, not "lifeless". And never recommend a loudness for its own sake - the right answer depends on the genre and the destination. What this page knows is what each destination will DO, which is a different and more useful thing to say. ## The sheet ```text PROGRAMME targets | spotify, youtube mode | album TRACKS t01 | -14.1 | -1.2 | 9.0 | 214 | the single ``` A track row is `id | integrated LUFS | true peak dBTP | loudness range LU | seconds | description`. Only the id and the integrated figure are needed; the rest widen what can be checked. `mode` is `track` or `album`. A FIELD THAT IS MISSING STAYS MISSING. A track with no true peak is checked for what it will sound like and not for whether it will clip, and the page says so - it is never treated as zero, because a zero would invent the exact fault this page looks for. ## The lanes - `plan` - picks the targets and says where one master cannot serve two - `check` - says what every destination does to every track, and which findings are faults - `peaks` - where it clips, and which of the three mechanisms the studio could not see - `album` - the same numbers under both normalisation modes, side by side - `deliver` - sorts findings into fix, accept and second master, which severity does not decide ## What it does not do It measures nothing and processes no audio. Every figure is arithmetic over numbers you supply, so it is only as good as the meter they came from.