When evaluating sound quality, audio engineers have traditionally relied on some explicit or implied comparison with a reference. Even if a reference signal is not formally used, any judgement about what is ‘good’ or ‘correct’ has traditionally been rooted in a concept of fidelity to some ideal. That ideal might be a memory of ‘the best I have ever heard,’ some naturally occurring version of the sound in question, or perhaps ‘what the engineer/producer heard in the control room.’ We assume that we know what correct sound reproduction is, and that high quality means getting as close to that goal as possible. This is all very well when we are mainly aiming at sound reproduction of something that actually happened, and when a canonical version exists. Audio products are increasingly interactive, though, and multiple versions can be rendered depending on user interaction. Extended reality systems make the user experience increasingly indeterminate. The concept of fidelity breaks down, and the engineer no longer has ultimate control over what is delivered to the listener. New paradigms for integrative evaluation of sound quality are needed.