Things are indeed now built on top of typechecker functions. And those functions indeed don't show you that much:
>> any-series?/
== ~#[frame! "any-series?" [value]]~ ; anti
But on the bright side, functions cache their symbols. So you're not completely in the dark about what a typeset represents.
Also: function dispatchers have reflectors now. So you can ask them questions. There's no reason the built-in typecheckers couldn't cook up an answer to TYPES OF...
>> types of any-series?/
== [text! block! ...<whatever>]
So look on the bright side, things aren't that bad.... 
Also... PARAMETER!
Another concept that may come into play here is PARAMETER!.
There's no MAKE for them yet. But we can get one out of the frame of a prebuilt function:
>> negate/
== ~#[frame!! [
number: #[parameter! [any-number? pair! money! time!]]
]]~ ; anti
>> negate.number
== #[parameter! [any-number? pair! money! time!]]
We can use it with match:
>> match frame.number 10
== 10
>> match frame.number <abcd>
== ~null~ ; anti
So that's something like a TYPESET!. It's not a set of bits, but an array of constraints augmented by some optimizations that speed up matching. I'm not sure how much of a role PARAMETER! will play, but it may fulfill some purposes people had in mind for typesets.