Just as I'm getting over my migraine from learning about Confidential Transactions at a high level we get another awesome proposal. When I first saw your request for compensation on the CoinJoin thread I thought you were just another crank and/or weirdo (The jury is still out.). I would very much like to see a proof of concept.
To be honest, I think the muted response speaks for itself. This stuff turns out to be less important than I thought. Someone will eventually re-implement the same things from the paper (maybe Blockstream or Monero guys?). Multiple independent implementations are a good way to do crypto, because then they can be compared and tested against each other.
I thought that CT represented the entire value in the mantissa, so isn't this a distinction without a difference?
Inputs do not have an exponent. The exponent is a property of the range proof, not of the values themselves. They work by scaling the basis the proof operates over.
I'm not 100% sure of the CT method, but it sounds like some information about the exponent is exposed to make the proofs shorter (you could keep it secret at a big cost). Maybe not the input magnitude, but the proof exponent range is public, and that selection, can itself give some information away. This is why CT is targeted at smaller 32-bit numbers.
Can CCT be used to encrypt other arbitrary information as well, or is it limited to transaction values?
Yes. There's a DH key exchange, which give a common secret through which you could share as much as you want.