Chocolate bar gives evidence for syntactic constraint

I bought myself a Cadbury's Boost (other chocolate bars are available, though none are as good if you want to ingest maximum calories). It describes itself thus:

2 x milk chocolate with caramel and biscuit filling bars

(It was a Boost Duo, OK? Don't judge me.)

Does that sound at all odd to you? Grammatically, I mean; it obviously sounds perfect in terms of content, though it does sell itself short, in my opinion. It's so much more than just 'caramel and biscuit'. But syntactically, it's really awkward.

Happily, the reason for its awkwardness can be attributed to the constraint that I work on. It's called the Final-Over-Final Constraint (FOFC), and it's about what phrases can be combined in what order. It specifically states that one order is not allowed, and it's the one instantiated in the description of my delicious chocolatey snack (although it's so many calories, it might have to be my delicious chocolatey tea).

This is the (partial) structure, as far as I can work it out:

[[milk chocolate [with caramel and biscuit filling]] bars]

Bars is the head of the phrase. It's at the end, as you can see. This means that it's a head-final phrase. FOFC states (basically) that a head-final phrase should not immediately 'dominate' (i.e. have as its immediate constituent) a head-initial phrase (that would be one where the head is at the start). The phrase milk chocolate with caramel and biscuit filling is just such a head-initial phrase, with chocolate as its head (we're not going to talk about milk now - it doesn't affect the argument). So we have precisely the relationship that FOFC doesn't like.

This type of construction is sometimes found: the quick-off-the-mark athlete, his out-of-the-blue question. These are marginal, for most people, and it's not a very productive construction: *a happy in his job employee is not good at all. The ones that are accepted are often taken to contain a lexicalised or Spelt-Out element - that is, the first part is not interpreted as having any internal structure, so any structural constraints don't apply to its parts, only to it as a whole, as if it was one word. The Boost description, I think we can agree, is definitely compositional (that is, it's built by the syntax, not interpreted as a single unit), so that explanation doesn't hold and we get a decidedly dodgy bundle of words.

In fact, this is such a mangled piece of syntax that there are at least two other reasons why this is bad. It leaves us hanging a long time before we get to the head, which we English speakers are none too keen on, and furthermore, the PP with caramel and biscuit filling modifies bars, not milk chocolate, so it should follow bars. Why it's where it is at all is beyond me. So we don't need FOFC to write this off as a bad job. But as we have FOFC for other reasons anyway, we can add it to the long list of Things Cadbury's Has Bungled.

(P.S. You may have noticed that Final-Over-Final Constraint violates itself. Its originators are quite proud of this, although it was accidental - the observation is attributed to Gertjan Postma. They note that all the best generalisations do so.)