A useful explanation I just found of how moving goal posts work with irreducible complexity comes from this review of Behe’s 2010 review of the literature: Behe's review in context, or what's the point?.
Behe constructs an elaborate apparatus for classifying mutations as “gain”, “modification”, or “loss” of what he calls a Functional Coded Element (FCT). The definition is skewed to make “gain” as difficult to prove as possible. The process needs to be understood at the molecular level, rather than simply in terms of phenotype expression. This enables him to dismiss as of unproven relevance the Lenski group’s famous demonstration of E.Coli acquiring the ability to metabolise citrate under anaerobic conditions. Moreover, advantageous removal of inhibition is treated as “loss”, but advantageous disruption of a function by IS duplication and insertion is classified as “modification”, rather than “gain”. Using these restrictive and asymmetric criteria, Behe classifies most sufficiently well-understood mutations in laboratory-bred bacteria as loss or modification, although he does recognise a few gains.
But there are numerous well-known counterexamples, many of them discussed in this review.
The next stage is rhetorical dismissal of such counterexamples. Here the strategies include limiting the search (ignoring the massive creative role of gene duplication and polyploidy in eukaryotes, and of horizontal transfer followed by selection in bacteria themselves), narrowing the criteria (new functions don’t count unless they can be demonstrated to arise from additions, rather than any other kinds of alterations, to the molecular machinery), and inventing additional constraints (creation of a new category, the FCT, classifying the process as a loss if either material or function is lost at any stage in the change being discussed, dismissing changes in function as mere transformations, rather than novelties). This stage switches the emphasis from what is possible in principle, to the demand that each case be demonstrated in practice, and fully analysed in detail, at the molecular level.
Finally, any counterexamples still surviving this moving of the goalposts and restricting and tilting of the playing field are dismissed as untypical, and therefore unimportant. Another leap of logic, as the present case shows. For even if losses (according to Behe’s criteria) outnumber gains, losses are in general unlikely to be dramatic without being lethal—there are some obvious well-known exceptions, such as the evolution of parasitism—whereas dramatic gains such as gene duplication, horizontal gene transfer, or polyploidy, can and do have the most profound effects imaginable.
You can read Behe’s paper here: