It's all about the natural causal chain.

Suppose I define a "creation algorithm" where you take a piece and subject it to a battery of complex transformations that themselves transform over time. It really would be a kind of musical encryption, and it would be very easy to make sure no human could discern the source material just by listening.

Now, if a piece is composed in the normal fashion, but some wit decides to make a name for himself by devising some such algorithm by which the newly written piece could be derived from an old one, there would still be no question of it being derivative, even though the means of derivation is presented. It wouldn't sound anything like the old piece, and we're assuming it appeared in the musical mind of its putative composer in a "mentally creative" way. The mere existence of a mathematical derivation process is not relevant. There is no natural causal connection between the first piece and the second piece.

However, if you can view the logs from the application that ran the creation algorithm, where each step and element of the transformation is executed, you will quickly come to: yes, THIS piece was hereby turned into THIS piece in a "computationally uncreative" way. The natural causal connections are there to be seen in the program execution. Presumably that's the kind of "yes" that awards full royalties.

This is related to but not exactly the same as the concept of "access". It's access in action.

Say Jack writes a song, puts it in a safe, and gives Jill the key. Then Jill publishes a song that's note-for-note identical to Jack's. She certainly had access, but if a microscopic examination of the lock shows it has never been opened, that basis for an infringement claim is moot. She may have had access, but she did use that access in a way that allowed the information inside the safe to flow outside.