The overall plan ....

Here is a rough synopsis:

(1) For broadly Davidsonian reasons, as well as others, we can roughly identify our 'conceptual scheme' with the conversation we are presently engaged in, and since we identify other things as conversations only if we can translate them, we end up with, in a rather ragged way, only one conceptual scheme. If we think it might have inconsistencies, we can try to find them & eliminate them. We have no intelligible way of speculating about what kind of language game this process would result in, because we would need to be able to translate it into the language we use now in order to indulge this speculation; and this is something we cannot, by definition, do.

(2) Although synthetic a priori beliefs may be elusive, synthetic a priori statements are easy to construct, as we can make general statements about the possibility of language: 'We can speak' is either true or not a statement. We need to be able to make assertions, because to deny this is to make an assertion. If we can make assertions, then 'We can speak' has the corrollary 'We can tell the truth'.

(3) Generic theories of truth (of how to speak) are blocked by the open question argument. Statements that it isn't possible to know we're telling the truth are self-refuting. So we must be able to know we're telling the truth without being able to say why, in order for language to work: and since we can assert the latter a priori, the former must be the case.

(4) The structure of fundamental arguments now becomes recursive: we can never argue 'X must be true because it is a theorem of my theory of truth', and we certainly can't argue 'X must be true because it isn't possible to tell the truth with certainty'. The only alternative is the transcendental argument: 'X must be true or it would not be possible to tell the truth'.

(5) Since 'We can speak' is an empirical statement, it is possible to construct fundamental empirical arguments transcendentally: Either Empirical(X) is true or it is not possible to make empirical statements. This is the bridge from reason to the world.

(6) While it may be possible to construct a transcendental argument (as in 5) for the reliability of the senses, this would not rescue traditional sensory empiricist epistemology - it would show that the reliability of the senses depended upon the reliability of empirical statements, and not vice versa.

Some stages obviously need elucidation (!) and there are lots of interesting consequences and further developments...

But the main elements are all here.