That's my best, "simple guy" understanding of this whole issue. Now I hope a real technical person can jump in and add to this, so this internal headroom issue can be "put to bed"
Actually it is pretty simple for the part we should understand:
Wikipedia: "Floating point""... whereas a floating-point representation ... with seven decimal digits could ... represent 1.234567, 123456.7, 0.00001234567, 1234567000000000, and so on. The floating-point format needs slightly more storage ... so when stored in the same space, floating-point numbers achieve their greater range at the expense of precision."
See what they do? It's all just numbers. If we leave a certain (mathematical) working range they shift the decimal point and continue calculating as before, just with another exponent.
Regarding the "precision" argument people can discuss if floating point or fixed point is better for audio, but there is no discussion about the headroom of floating point calculation. Of course there is a ceiling somewhere but it is very far away. Since floating point arithmetics permanently shift the head up and down in such a big space, it makes no practical sense to talk about the room above the head.
---
Maybe the following is a better explanation for musicians:
You have a small keyboard with 32 keys. This means you can play 3 octaves, ok? If you try to play higher you will miss the keys and finally fall off the stool because there are no keys up there. But wait, this is a "floating point keyboard"! You simply press a button and shift the octave range up. Amazing, you play as before, but it sounds higher. If you don't need more than 32 keys simultaneously, with octave transposition you can cover much more octaves than there are keys.
Compared to floating point arithmetic, the 32 keys are the available decimal digits and the octave transposition button is the exponent. The higher you set the exponent, the higher goes the sound, but the keys remain as they are.
Now, who would say, that a g''' is not playable on a small electronic keyboard because there is no key for it? Just shift the octave. And who says that Logic can't handle +300 dB because there is some miraculous limit at zero or +6 dB? This is not the case, Logic shifts the arithmetical working range, so to say. Not even the output channel clips - what we hear is the the clipping of the D/A conversion.
---
Here is an
interesting video that shows a practical usage of floating point calculation. We cannot do this in Logic because we don't have this export option. But Logic does it itself: When we freeze tracks, Logic writes a 32 Bit floating-point file to disk which is used for playback. If we unfreeze, Logic deletes this file and starts to calculate the track in realtime again.
---
As usual we must be aware that this is a technical discussion. We should not mix all in the red because we can. Some plugins do not like that. We don't get optical feedback about the levels. Finally we have to bring all down anyway. And more general: this is bad behavior
😉