In the past I've expressed the belief that you don't have to worry about keeping your tracks out of the red, as long as you pull everything down prior to going into the master bus. That was based on the belief that there is no serious harm done to floating-point data that occasionally goes over 0dB, as long as you ultimately provide adequate headroom before hitting the master limiter. Everybody knows floating-point audio doesn't clip, right?
 
I was wrong about that. I feel bad about having made that assertion and misleading others, so this post is to make amends.
 
Turns out, there are a surprising number of plugins that distort in an unpleasant way with input > 0dB, or that generate > 0dB internally. Testing has revealed that the list includes some of my favorite plugins, from highly-respected vendors that have a reputation for great sound quality.
 
My epiphany occurred a few months ago, when I noticed some ugly popcorn-type distortion on a track and set about hunting it down. It turned out to be a SampleTank instrument, and the distortion was occurring within the plugin itself. Turning its internal volume down solved the problem. However, the instrument was now too quiet, so I had to turn every other track down, too, via their Trim/Gain sliders. (A pain, because it meant re-doing every track compressor.)
 
Had my old assumption been true, there would have been no change to the overall sound of the mix. To my surprise, the difference was huge. The whole mix had become clearer.
 
There had been an embarrassing number of little red flags scattered throughout that project. Several plugins had been introducing subtle distortion as a result. Aside from the SampleTank track, none of these distortions was individually noticeable, but they had a cumulative effect. Even then, it was not obvious there was a problem - until it went away.
 
Nowadays, I make a point of maintaining adequate headroom from the get-go. Sorry for having promulgated wrong information.
 

I have a brand new, for me, Kontakt 5 patch, Upright Piano Modern Preacher, that hits the "red" with moderate velocity mapping in the MIDI source. I came upon the circumstance just yesterday.
 
Yuck.
 
 
My suggestion: mix with the powered monitors switched to "on". 
 
:-)

Hard Top. 103% if you are lucky on a computer. 105% maybe, but probably not.
 

In a dense mix, one of the single most challenging things to get right can be gain staging.Especially when you have the occasional peak transient on an otherwise great track.Let it go or lower the whole track ? Edit midi velocity? Apply some clip compression in only a few places? Volume automation? Lots of approaches, but it isn't always as easy as it seems.
 
Pianos can be some of the toughest IMO. But they aren't alone. 24 bit seems to be a bit less forgiving, but that might be a misconception on my part.

happiness is clean audio.

Starise
24 bit seems to be a bit less forgiving, but that might be a misconception on my part.

You're right, because when we say "24-bit" we're implying integer values. For integer (as opposed to floating-point) numbers, there simply isn't anything beyond 0 dB. That's why it's so important to not drive your audio interface into the red, because it knows nothing of decimal points and therefore has an absolute upper limit that literally can't be crossed.
 
Floating-point, OTOH, automatically adjusts to whatever size number you want to stick in there. It just shifts the decimal point as needed. It can therefore accommodate very large (or very small) numbers without running out of bits. You do lose some accuracy as some of the least-significant decimal places are chopped off, but this loss of accuracy is irrelevant in audio, so in theory 32-bit floats can go over zero with impunity.
 
It's a wonderful innovation, using floating-point data in DAWs. However, it breeds sloppiness because it allows us to get away with practices that wouldn't have been acceptable in the pre-digital age. And presuming everything's OK can occasionally bite ya in the arse.
 

My suggestion: mix with the powered monitors switched to "on".

Good advice!
However, it was because I had the speakers off (listening on cans) that I initially noticed the crackling that headed me to this discovery.

Bit, some plugs are designed to start distorting before you reach zero. They model overloading before the signal is clipping.

bitflipper
in theory 32-bit floats can go over zero with impunity.

You can exceed zero and never clip - this much is always true (as long as you appropriately attenuate before a conversion back to fixed point format happens, of course).
 
On the other hand, I think I'd feel cheated if a plug-in that purported to emulate analog gear didn't behave differently when processing signals that would be considered too hot in the analog world.  If that plug-in doesn't do something different when being fed too-hot data like the real gear, then it's not behaving as advertised.

This explains the popularity of all those tape emulators. 
 
You're supposed to hit the red hard with those things.

I guess this is why the ProChannel modules have clip lights?