[SOLVED] At what stage is the audio finally clipped?

Interesting result.
 
I presume one of the Bakers can confirm your findings definitively, but I assume this is due to the fact that the audio is routed internally at a higher bit depth than the actual bit depth of the project so there is additional headroom until the very last output stage.
 
AMB

rogeriodec

ampfixer
If it doesn't clip within the DAW, where does it clip? I'm feeling pretty stupid. 

After this my study and after the collaborations in this topic, I understood, in summary, that the audio clipping occurs effectively in 2 moments:
1) In the last step of the last BUS, where the final audio file will be generated;
2) Or, eventually, within some VST placed on some intermediate track or bus, which does not know how to deal with excessive volume.

There is one other, less likely case: if a user has set the render bit depth to 24 or 16, any intermediate clip/track bounced or frozen could clip. The default is 32 bits, which behaves like all internal calculations, allowing large level changes without clipping, and with great dynamic range. The render bit depth is configurable in Edit > Preferences > File - Audio Data : File Bit Depth : Render Bit Depth.

Clipping occurs when audio signals represented as integers exceed the maximum or minimum possible value for an integer. 24 bit integers can store a much larger range of numbers than 16 bit integers, but can still get maxed out.
 
Your audio interface works with integers, both going in and coming out of the DAW. However, internally a DAW works with floating point numbers. The conversion happens when importing/recording or exporting audio. This includes reading/writing wav files stored in the audio folder. wav files are stored in integer format.
 
Floating point numbers can represent a MUCH larger range of values. Once audio data is within the DAW software, it is extremely difficult (effectively impossible) to get clipping.
 
However, as others have noted, some plugins may react badly to large floating point numbers. I'm not sure why - maybe they make some assumptions about the size of the incoming signal. The meters in Sonar must make similar assumptions. So best to keep the signal amplitude in the sane range, but behind the scene, insanity is tolerated!
 
Cheers, Peter.
p.s. It just occurred to me, why don't audio interfaces convert analog audio input to floating point numbers? Probably because the typical analog/digital conversion chips produce integers (16/24/32 bit).