There is a tidy argument going around: 32-bit float processing has made gain staging obsolete. You cannot clip the mix bus anymore, so why bother setting levels at all? Drag the faders wherever, the float engine sorts it out.
The first half of that argument is correct. Float genuinely does eliminate internal clipping, and the reason is worth understanding properly. The second half is wrong, and it is wrong because of the exact same property that makes the first half true. Float protects the number. It does not protect the plugin the number passes through.
Why float eliminates internal clipping
Fixed-point audio, the 16- or 24-bit integer kind, represents amplitude on a fixed ladder of equally spaced steps. There is a hard top rung at 0 dBFS. Cross it and you have two options, both bad: wrap the value around or clamp it flat. Either way the information above the rung is gone.
Floating point stores a number the way scientific notation does. There is a mantissa, the significant digits, and an exponent, the scale. In IEEE 754 single precision that breaks down as 1 sign bit, 8 exponent bits, and 23 stored fraction bits. Add the implicit leading 1 and you get 24 bits of effective precision.
The decisive detail is that the exponent is free to move, and the 24-bit mantissa keeps its relative precision no matter where the exponent sits. A signal at -60 dBFS carries almost the same precision as the same signal at -1 dBFS. The exponent just slides the scale up or down.
So when a float signal goes "over 0 dBFS", nothing gets clamped. The exponent grows, the mantissa still describes the waveform shape to 24-bit accuracy, and a later gain reduction slides the exponent back down with nothing lost. That is why a 32-bit float file that looks visually clipped can be pulled back with a gain slider and sound clean. The often-quoted dynamic range of roughly 1528 dB is just the consequence of how far that exponent can travel.
This part of the story is real. Inside a float DAW you can summon a 200 dB peak on the summing bus and it will be carried faithfully.
Why that does not save your mix
Here is the boundary the obsolete-gain-staging argument quietly steps over. Recoverability only applies to a sample value passing through linear operations: gain changes, summing, panning. The moment the signal enters a nonlinear process, the input level changes the output in a way no fader move can undo.
Saturation generates harmonics whose amount and character depend on input level. A compressor's gain reduction is a function of how far the signal sits above its threshold. Analog-modeled EQs, preamps, and tape emulations recreate hardware whose tone shifts with drive. As Sound on Sound puts it, analog-modeling plugins "often do a great job of mimicking the sonic behaviour of their hardware equivalent in its normal operating range, but start to fall down when you push them outside that range."
Once a saturator has added the harmonics for a signal peaking at -3 dBFS, pulling the channel fader down afterwards lowers the volume of the distorted signal. It does not remove the distortion. Float gave you back the number. It cannot give you back the behavior.
That is the whole point. Gain staging in a modern DAW is no longer a clip-avoidance task. It is a tone decision. The question stopped being "will this clip" and became "is this plugin getting the level it was designed for".
The calibration question
Analog gear has a nominal operating level: 0 VU, which corresponds to +4 dBu in professional equipment. A plugin developer modeling that gear has to choose where 0 VU sits on the digital scale.
The most cited choice is -18 dBFS, which leaves a console-like amount of headroom above the nominal level. Send a signal peaking at -3 dBFS into a plugin calibrated for an -18 dBFS average and you are driving it roughly 15 dB hotter than intended. For a clean digital EQ that may not matter. For a tape emulation it is the difference between a hint of glue and an obvious effect.
But -18 dBFS is a convention, not a universal spec. European EBU practice (R68) maps 0 dBu to -18 dBFS. US practice (SMPTE RP155) maps +4 dBu to -20 dBFS instead. The two standards reflect different headroom philosophies, not a mistake on either side. And some plugins make the choice explicit: Universal Audio's tape machine emulations expose a selectable calibration level precisely because different tape machines and formulations behaved differently and there is no single right answer.
Most plugins never document their internal reference at all. So the honest takeaway is not "set everything to -18 dBFS." It is narrower: the plugin has a level it sounds best at, that level is roughly console-like rather than near 0 dBFS, and your job is to land the signal in that neighbourhood and then trust your ears.
Where the hard ceiling comes back
Float forgiveness also has an expiry date. The summing bus, the channel path, plugin-to-plugin transfer: all floating point, all forgiving. But a D/A converter is a fixed-point device, and an exported 16- or 24-bit WAV is a fixed-point file. At both of those points 0 dBFS is once again an absolute wall.
A mix that "sounds fine" running 6 dB hot inside the DAW will clip the instant it is rendered to a delivery format or sent to a converter with nothing trimming it down first. The float-makes-gain-staging-obsolete claim fails even on its own clipping-focused terms. It is only true for the middle of the chain, never the ends.
What to actually do tonight
The reframe is the core of this. Stop thinking of gain staging as clip-avoidance and start thinking of it as feeding each plugin the level it expects.
A few concrete moves that follow from that:
Stage at the input, not the output. Insert plugins typically receive signal pre-fader, so the channel fader does not change what the plugin hears. Use a gain or trim utility before the nonlinear plugin to set its drive. The fader stays free for balance.
Aim for console-like levels. Average levels around -18 dBFS, peaks roughly -10 to -6 dBFS. That lands most analog-modeled plugins near their sweet spot without you needing to know each one's exact internal reference.
Use the plugin's own controls. Many analog-modeled plugins have input or drive and output or trim knobs for exactly this reason. If a saturator sounds too aggressive, the fix is the input knob, not the channel fader.
Trust ears over meters for nonlinear gear. The -18 dBFS target gets you into the room. Whether a specific compressor or tape plugin sounds right is decided by listening, because its real calibration is usually undocumented.
Protect the ends of the chain. Leave several dB of headroom on the mix bus and put a limiter or trim before any export or converter stage.
There is a worthwhile side note here. Some experienced engineers believe DAW summing engines do not produce identical results when summing many very high-level signals, due to floating-point rounding. Sound on Sound presents this as a held belief rather than a measured fact, and modern DAWs commonly sum in 64-bit double precision anyway, which makes the concern smaller still. It is not a reason to panic. It is one more small argument for not running everything hot.
If someone tells you gain staging died with 32-bit float, the one-sentence reply is the same line from the top of this article. Float protects the number, not the plugin.