Well, what I can wildly guess for now is that at least some kind of “interfacing operation” is virtually conceptually necessary (since CI/SM linking is the whole function of Syntax to begin with). As far as I can see there is no a priori reason for us to believe that the PF-interfacing and the LF-interfacing should exactly correspond in timing (as the Transfer theory assumes), that they are designed to occur more than once in a given derivation (as the multiple Transfer theory assumes), that they are designed to target a particular domain indicated by a designated lexical item called a phase head (as the Chomskyan phase theory assumes; compare it with Uriagereka’s multiple S-O story), or that the domain subjected to interfacing becomes invisible for later operations (as the PIC says, but see Boskovic), so all of them are intriguing open empirical questions, but at least I could have convinced myself that we might be able to learn a great more about human language by studying how the interfacing works.

Thus, my present position for phases is this: once we allow ourselves to assume that interfacing is virtually conceptually necessary, we would be immediately interested in how far we can go only with Merge and Interfacing, and Interfacing as a multiply applicable phase-head-driven `sound’-`meaning’-uniform computational load-reducing PIC-obeying operation (as people nowadays assume) is just one possible hypothesis on how Interfacing works, quite possibly subject to revision. The absence of strong empirical evidence might be already suggesting that this particular hypothesis may be wrong, but all of these plausibility arguments might be valuable, since we
are anyway availing ourselves of Interfacing and SMT, we want to have the most conceptually prefererable hypothesis of Interfacing, thus there may be some reasons for us to bet our money on it (but always with saving some reservation).