Well, one reason we might get stuck on a “suboptimal” language is network effects outside of the criteria we use to compare languages: what language does an organization already have experience with, does the (only) current compiler support the target hardware and system, it’s to late now to write the Linux kernel in Haskell, etc.

However, I agree with you. Despite the time I’ve invested in learning Haskell, I’ve personally decided to spend my time developing in other languages. If Haskell is the magic wand right there in front of us that will give us programs without flaws, how come almost nobody has picked it up and shown us working programs that took less development effort and fewer flaws.

We could compare John Wiegley’s ledger (C++) to the similar hledger (Haskell), but hledger isn’t the same program as ledger it has different features, I have no idea how long it took to write, how many flaws it has versus the original, how different the developer skills were, how much time and defects were saved by it being a redesign, etc. And these are small single developer efforts (now with numerous contributors). From everything I can gather, they are both excellent. So in this case there is no compelling advantage demonstrated for Haskell, and this supports your point.