Fat, punchy, loud — pick three

The brief arrived as two words: make it move, make it fat. Twenty tracks, different genres, all headed for reels — played on phones, at arm’s length, against every other loud thing in the feed.

It took four failed attempts to understand that those are two different physical things, and that confusing them is why most loud masters are flat.

Movement is gain changing in tempo. The kick hits, everything dips, and it recovers before the next kick. Kill it three ways: attack too fast (the hit gets shaved), release too slow (the gain never comes back between hits — sausage), or simply pressing everything down hard (no movement, just quieter).

Fat is density between the hits — harmonics, sustain, the room coming up, low-mids left alone.

The first is timing. The second is saturation. One compressor, run the usual way, gives you either at the cost of the other.

Asked to be loud, it made quiet

Our first pass was the classic move: a good levelling compressor on the bus, run hard. Across twenty tracks the “gentle” setting came out five dB quieter than the raw mixes. The “slam” setting came out loud — and flat as a table, every hit the same size as the bed under it.

Nothing was broken. Deep serial compression buys sustain and pays with transients, and once you level-match against a peak ceiling, paying with transients means paying with loudness. The compressor was never the wrong tool. It was the only tool, and it was doing two jobs that contradict.

The fix was a routing change, not a setting

The single biggest improvement of the whole project was turning one knob down: the saturator’s mix.

Run full-wet, the compressed path is the signal, so its dead transients are your transients. Run parallel — 60 to 85 percent — the dry path carries the hit intact and the wet path carries the weight underneath it. You stop trading punch for density and get both.

Measured at the same loudness, the punch metric we track went up by two and a half dB the moment the chain went parallel. Fat and moving, which the serial version could not do at any setting.

This is why Ferment Charge has a Mix knob on the front panel instead of buried in a menu: parallel is not a trick, it is the correct default for bus work.

Clip first, then limit

The back end of a loud master is two devices in a strict order.

The clipper has no memory. It shaves the top off a hit and has forgotten it by the next sample — so it cannot pump, no matter how hard it works. Taking several dB off drum peaks here is normal, and nobody hears it as compression because it isn’t any.

The limiter then only has to manage what’s left. Two to four dB, released musically, instead of ducking the whole mix around every kick.

Reverse the order and the limiter meets every spike at full height — which is exactly the pumping that reads as “flat”. Our rule: the clipper’s ceiling sits about half a dB above the limiter’s. The clipper does the fast work; the limiter rounds off what the knee lets through. That half-dB gap ships as the default relationship between Ferment Clip and Ferment Limit.

The track that blew up was the emptiest one

Somewhere past “pleasantly dense” a master starts to choke — and the track that choked first surprised us. Not the dubstep with the most bass. The dub techno: a bare sustained sub and almost nothing in the mids.

That’s the mechanism. Saturating a dominant sub throws distortion products all across the midrange, and in a sparse mix there is nothing to hide them behind. The dubstep swallowed the same distortion whole — its mids are full of noise and stabs that mask everything.

So the failure is never “too much saturation”. It is too much saturation on material that cannot hide it. The fix, when the sub dominates: a gentler saturation voice, keep the bass out of the compressor’s ears, and pull the parallel blend drier. We found it by switching one stage off at a time at matched loudness — the culprit was obvious in one pass.

Loudness is not level

Past the target, more gain stopped helping. Two things still did.

Presence. The ear judges loudness mostly by the 2–5 kHz region, so one dB of presence and a slightly brighter tilt read as “louder” without adding a single dB of peak. Our loudest-sounding setting is not our highest-measuring one.

And the codec: encoded audio overshoots the ceiling you set, because the decoder reconstructs peaks the file never had. We built a loop that lowered the ceiling until the decoded file behaved. It never converged — a lower ceiling at the same loudness packs the waveform denser, and denser overshoots harder. One track spiralled five dB down before we stopped it. Set the ceiling, accept about a dB of codec overshoot, and spend your attention somewhere it pays.

Two masters, one track

We ship every track twice, and the choice between them is not taste — it is the delivery channel.

Studio is the conservative one: for monitors, big systems, streaming under normalisation — places that supply their own scale, where the breathing master wins. Reel is a dB hotter with a harder ceiling and a brighter tilt: for the phone in a noisy room, which supplies nothing.

The moment we trusted the split: the Reel master held up on serious studio monitors too. The aggression was structural — parallel density, real dynamics — not a psychoacoustic trick that collapses on good speakers.

The preset that failed

At the end we ran the experiment that explains why Ferment exists. We took the exact chain, the exact knob positions that had just mastered twenty tracks, and bounced a track through it in a DAW like any preset.

It came out five to eleven dB quieter, with the mix’s problems intact.

The plugins were bit-identical. What was missing was everything around them: gain staged into the chain per track, the sub treated only when the sub dominated, the tone shelf sized to what this mix measured, the width pulled in only when mono folding actually lost level. A preset is the same numbers for every track. The work was never the numbers — it was the measuring.

That gap is the product. Ferment Percept is the measuring, as an instrument you can read. Ferment Master is the chain from this article with the measuring built in: press Learn, play the loudest section, and ten seconds later it has staged, voiced, tilted and levelled the track from what it heard — onto ordinary knobs you can disagree with.

The article you just read is what the button does.