LUFS, true peak, and the platform second pass

The number on the meter is not the whole trip

You print -14 LUFS integrated, short-term peaks behave, and the reference track finally sits in the same zip code as yours. Then the upload hits Spotify or Apple Music and the vocal gets a little smaller, the snare tick loses air, or the whole thing walks a half dB away from where you left it. That is not mysticism. It is a second pass you did not budget for.

What is actually happening

LUFS tells you how loud the program is over time in a way that correlates with human loudness perception. Streaming platforms use integrated loudness targets so catalogs do not jump like a badly programmed radio station from the 1990s.

True peak is the other half of the handshake. Sample peaks can lie. Inter-sample peaks happen when the DAC or the codec reconstructs a waveform that exceeded 0 dBFS between the dots on your grid. A limiter with true-peak awareness tries to bound those overs, but the model is not identical across every encoder, player, or car stereo DSP.

Then the platform applies its own chain: loudness normalization, sometimes additional peak control, and lossy encoding that smears transients depending on bitrate and joint stereo decisions. Your file is not the version listeners hear in the abstract. It is an input to a black-ish box that is allowed to protect its own UX.

Why it matters

| Stakeholder | What they optimize for | |-------------|-------------------------| | You (mixer/mastering) | Translation, punch, emotion, repeatability | | Aggregators / stores | Consistent perceived loudness, fewer user complaints, predictable peaks on cheap earbuds | | Listeners | Not touching the volume knob; not getting ear fatigue in one track |

When incentives line up, everyone wins. When they do not, the platform still wins because it can nudge your audio until the complaint graph flattens.

What breaks

• Mastering to the meter, not the encode. A print that kisses -1.0 dBTP in one tool can read differently elsewhere. If your limiter is not oversampling enough for the material, you are mastering a JPEG preview.
• Genre-specific transient density. Short, high-energy bursts eat more headroom in AAC than a pad wash at the same short-term LUFS. Two masters at -14 LUFS integrated do not necessarily survive the codec the same way.
• Client revisions after loudness pass. The classic trap: someone bumps a vocal 0.7 dB on the “final” and nobody re-checks TP before upload. Looks great in a demo. Then you try exporting stems.

What to do next

• Pick one loudness meter you trust for integrated LUFS and one true-peak path you actually use at the end of chain. Log which preset (oversampling, ISP mode) was on when you signed off.
• After limiting, run a lossy encode round-trip on a 30–60 second loop that includes your loudest section and your most fragile transient. A/B at matched loudness, not matched fader position.
• Document a house ceiling that is boringly safe on true peak for your genre (numbers vary; the habit does not).
• Ask mastering engineers or aggregators which measurement spec they guarantee for delivery (LUFS algorithm variant, true-peak mode). Misalignment between supplier and platform is cheaper to fix before release than after.

Bottom line

Integrated loudness is the table stakes bet. True peak and the platform codec pass are where the house still has an edge.

One thing to try this week

Take a finished master, encode it to 256 kbps AAC (or the closest proxy your tools offer), decode back to WAV, null against the pre-encode print at matched alignment, and listen to the residual for 60 seconds. You will hear what the second pass is stealing or smearing before any marketing department names it “clarity enhancement.”