That was in the other article, the one I didn't comment on at all.
There are two reasons I didn't comment on it. First, I don't actually understand what the author means by things like "only phrases that satisfy typing judgements have meanings" and "a typing judgement is an assertion that the meaning of a phrase possesses some property". I can kinda-sorta map this onto my intuitions about compile-time and run-time typing, but when I tried to actually drill down into the definitions of things like "phrase" and "typing judgement" I got lost in category theory. Which brings me to the second reason, which is that this is very, very deep rabbit hole. To do it justice I'd have to write a whole blog post at the very least. I could probably write a whole book about it. But here's my best shot at an HN-comment-sized reply:
I've always been skeptical of the whole static typing enterprise because they sweep certain practical issues under the rug of trivial examples. The fundamental problem is that as soon as you start to do arithmetic you run headlong into Godel's theorem. If you can do arithmetic, you can build a Turing machine. So your type system will either reject correct code, accept incorrect code, or make you wait forever for an answer. Pick your poison.
Now, this might not matter in practice. We manage to get a lot of practical stuff done on computers despite the halting problem. But in fact it does turn out to matter in practice because in practice the way typed languages treat arithmetic generally imposes a huge cognitive load on the programmer.
To cite but one example: nearly every typed programming language has a type called Int. It is generally implied (though rarely actually stated) that it is meant to represent a mathematical integer (hence the name). But computers are finite state machines which cannot actually represent mathematical integers. So in practice Int generally means something like Int32 or UInt64 or maybe, if you're lucky, a generic numerical type that will automatically grow into a bignum rather than overflow or (if you're unlucky) wrap around if you try to add the wrong values.
Sometimes these details matter and sometimes they don't. Sometimes when I'm adding things I really care about how it's done under the hood. Maybe I'm writing something mission-critical that can absolutely not fail, or maybe I'm writing an inner loop that has to run as fast as possible at the cost of possibly failing on occasion. But sometimes I don't care about any of this and I just want to add, say, two elliptic curve points by writing P1+P2 rather than EllipticCurveAdd(P1,P2) -- or was that EdwardsCurveAdd(P1,P2)? or Curve25519donnaPointsAdd(P1,P2)? -- and I don't care if it takes a few millisecond because I'm just noodling around with something. If I have to take even 30 seconds to read the docs to remember what the name of the function is that adds elliptic curve points, I've lost. It doesn't matter so much if I only have to do it once, but in practice this sort of cognitive load infects every nook and cranny of a real program. Thirty seconds might not matter much if you only have to do it once. But if you have to do it all the time it can be the difference between getting your code to run in a week and getting it to run in ten minutes. And God help you if you should ever have to make a change to an existing code base.
Those are the sorts of things I care about. Those concerns seem worlds away from the kinds of things type theorists care about.