Finagle My Song · Blog
Why your -9 LUFS master comes out at -11
You set the target to -9. The chain runs. You measure the file that comes out and it reads -11.
Nothing went wrong. That's the system working.
The last stage gets the final say
Loudness isn't the last thing that happens to your master. True-peak safety is.
Once the renorm has pushed the track to your target, something still has to stop the peaks distorting on playback. That stage isn't optional and it isn't polite about it. If hitting -9 has pushed inter-sample peaks past the ceiling, the limiter pulls them back down, and gain reduction applied a few hundred times across a track shows up in the integrated measurement.
So the number you asked for and the number you get are two different things, and the gap widens the harder you push.
Why dense material loses the most
This doesn't hit every track equally.
A dynamic mix arrives at -9 with room to spare. Peaks are occasional, the limiter barely engages, and the file measures roughly what you asked for.
Dense material is the opposite. It's already compressed, the crest factor is low, and there's almost nothing between the average level and the ceiling. Ask that track for -9 and the limiter works constantly to keep the peaks legal. On the tracks where we've measured it the claw-back runs to about 2 dB, which lands a -9 target somewhere around -11.
Here's the part worth sitting with. A -11 target on that same track would have measured about -11 too, and it would have got there with a fraction of the limiting.
Same loudness. Less damage. The louder setting bought nothing.
You can't hear this without measuring it
The reason this goes unnoticed for years is that nobody checks.
Your DAW shows you the target you dialled in. Your export dialog shows you the target you dialled in. Unless you run the finished file back through a meter, the number you remember is the number you asked for, not the one you shipped.
This is why our chain measures the true integrated loudness of the file after every stage has run, including true-peak safety, and reports that number rather than the target. It's the only figure that describes the file you actually have. A target is a request. The measurement is the answer.
Then streaming turns it down anyway
There's a second loss stacked on the first.
Spotify normalises to around -14 LUFS. So do YouTube and Tidal. Apple Music sits nearer -16. Your -11 and your -9 both get turned down to the same playback level, so on streaming the extra limiting you paid for buys you no volume at all. It just arrives more squashed.
That's the trap in full. You push past the point where loudness stops accruing, you take on limiting damage to get there, and then the platform flattens the difference you were chasing.
Loudness genuinely does matter on destinations that don't normalise. A downloaded file, Bandcamp, a DJ set off local files, a car stereo playing from a USB stick. Even there, -9 is already loud. The gap between -9 and -6 is not what's holding your track back.
What actually survives being turned down
Something does carry through normalisation, and it isn't the LUFS number.
Saturation survives. Upper-mid presence survives. Transient shaping survives. When a master still hits hard after Spotify has turned it down 5 dB, that's what you're hearing. The density stays; the level was never the thing doing the work.
So when a track feels weak next to a reference, reach for grit and presence before you reach for a louder target. One of those still exists after normalisation.
The practical version
Master to -11 unless you've got a specific reason not to.
Then measure what came out. If your file lands meaningfully below the target you set, your chain is fighting itself, and the fix is to ask for less rather than push harder.
If you want it to hit harder after that, add saturation or presence and leave the target where it is. Check the two against each other at matched volume, because an unmatched A/B just tells you which one is louder.
That last part is worth doing properly. Turn the louder version down until the two measure the same, then listen. Most of what a loudness increase seems to improve turns out to be the increase itself, and it disappears the moment you level-match.
This is baked into the tool I built. Finagle My Song measures your track, recommends a target derived from its own loudness and dynamics rather than a blanket number, reports the true output loudness after the full chain, and lets you A/B every version against your original at matched level. Analysing and listening are free, no account.
Set a number you can actually keep. Chasing one you can't just costs you the dynamics you had.