Recently, Fujisaki and Okamoto provided a revised version of the Fujisaki-Okamoto transformation [1], a generic transformation for achieving IND-CCA2 security in the Random Oracle model. This new version is slightly different than the original [2] and the security proof does not define a knowledge extractor (the original version actually proved that the transformation satisfies the Plaintext Awareness (PA) notion and used the fact that PA $\Rightarrow$ IND-CCA2 in the random oracle model). Instead, they directly provide an algorithm for the decryption oracle. When describing the decryption oracle construction, they stress that it does not require the secret key (in a similar way than the original version), and it only uses the information from previous random oracle queries.
Is this characteristic (i.e., not using the secret key) relevant in general for proofs for the IND-CCA2 notion, or is it simply something inherited in this particular case from the original transformation, aimed at the Plaintext Awareness notion? In other words, is there any inherent drawback in using the secret key in the decryption oracle when facing a security proof for IND-CCA2? I know that, usually, proofs are constructed as reductions to hard problems, which may force to define the decryption oracle without knowledge of the secret key. However, this does not seem to be the case.
I think I may be overthinking the problem, and that the answer may be a simple "No".
Update: I added some possible explanations in the comments. I'm starting to lean towards the answer "It is important in this particular proof, but not in general".
References:
[1] Fujisaki, E., & Okamoto, T. (2013). Secure integration of asymmetric and symmetric encryption schemes. Journal of Cryptology, 26(1), 80-101.
[2] Fujisaki, E., & Okamoto, T. (1999). Secure integration of asymmetric and symmetric encryption schemes. In Crypto (Vol. 99, No. 32, pp. 537-554). [link]