There’s a bit of confusion here. You don’t record in 32 bit. The only reasonable choices are 16 or 24. None of us has hardware that is capable of 32 bits.
32 bits floating is Audacity’s internal format. Once your recording lands in Audacity, it’s up-sampled to 32 bit automatically.
2. Record in 32, export as 24, repair, import, and edit, then export as 16 for CDs.
1st option is dithering once, 2nd one dithers twice. Right? Does it really matter?
You record in 24 bit, at a sensible sampling frequency of your choice.
After recording, I immediately store an archival copy at 24 bit, 44.1 or higher if needed.
I import the archival copy in whatever I work with atm. Most audio editors and DAW’s are non-destructive and if used properly, will leave the archive copy alone after import. In Audacity, the import is cut up in small fragments (.au files) and there’s an .aup text file that contains the logic to stitch them back together. All in 32 bit.
Whenever I get to a result, I make a 1.0 copy, again in 24 bits.
Finally, when it’s finished, I export the final version in 24 bit and the “output” version in 16 bits, 44.1 or 48. Or even mp3 if it’s just for background theater stuff that’s played on an mp3 player anyway.
I never use FLAC, for compatibility reasons. If needed, I zip wavs. The result isn’t as small as with FLAC, but usually it avoids cutting the wav file in pieces for mail, fi.
Dithering only plays when down sampling. Having a “universal” archival format, avoids downsampling until it can’t be avoided. In te last step. And when I’m there, I don’t worry about dithering.
There’s a comparison out there on the net, with beautiful graphics of DA’s and dithering and phase behaviour and linearity and stuff like that. The only thing that it makes clear, is that you shouldn’t worry about it, because the audible effect is minimal and far, far less than downsampling to mp3, fi.
Usually, I don’t even save Audacity projects at all, as I might continue working with the audio in Final Cut, fi.