A piece that felt enormous on the scoring stage can arrive on a streaming platform sounding polite. Nothing was broken in between. What happened is loudness normalization, and once you know how it works the fix is mostly a matter of deciding what you want to trade.
Loudness is measured in LUFS, a scale built to track how loud something seems rather than how high the peaks are. Broadcast settled on targets years ago: EBU R128 asks for about minus 23 LUFS across a program, and the American equivalent, ATSC A/85, sits at about minus 24. Streaming music platforms normalize to something nearer minus 14. Those are not small differences. They are the difference between a mix that keeps its dynamic range and one that has been pushed.
Normalization means the platform turns everything toward its target. A quiet piece gets turned up, a loud one gets turned down. So the loudness war has no prize at the end of it any more: squash a cue to sit at minus 8 and the platform simply attenuates it, and all you have left is the squashing. The piece that was compressed hardest arrives at the same perceived level as the one that was not, with less of its shape intact.
For film and television there is a second rule on top of that, which is that the mix is anchored to dialogue. Everything is balanced against the level of a voice, because a viewer sets the volume by whether they can hear what is being said. A cue is loud or soft relative to speech, not in absolute terms. On the stage you heard the cue at the level the room gave it. At home it has been placed under a dialogue anchor, through a dynamic range that the platform or the television may be narrowing further with a night mode nobody told you about.
So three things are usually happening at once. The program is normalized toward a target, the music sits beneath a dialogue anchor, and the playback device is applying its own processing. Each is reasonable. Together they take the top off anything that relied on sheer level for its impact.
What survives all of that is contrast rather than loudness. A cue that steps down before it steps up still lands, because the relative move is preserved when everything is scaled. Low end that is defined rather than merely large survives too, since a phone or a soundbar reproduces a shaped low end better than a broad one. And arrangement survives best of all: if the weight is coming from four instruments playing something that moves, it arrives. If it is coming from a wall of sustained sound at maximum level, normalization takes exactly that away.
It is worth metering what you are actually delivering rather than guessing. Any modern loudness meter will show integrated LUFS for the whole piece, short-term for the moment, and true peak. Mix to the spec you were given, check the integrated figure against the platform the piece is headed for, and leave headroom rather than spending it.
The short version: the platform decides the final level, not you, so spend your effort on contrast and arrangement, which the platform cannot flatten, instead of on level, which it will.
