It was certainly important for us to make sure developers were happy with the result. That's why our rollout included stuff like a phase where developers continued writing Skew but their changes automatically checked-in generated TS. This way, devs could see what it would look like in PR review and report issues. As for performance and runtime correctness, you're right that there are some gotchas with TS. We caught issues like the array destructuring one with instrumentation and strict monitoring.

I was definitely sad to see some features of Skew go. For example, operator overloading and integer types. But the move was ultimately a decision the whole team made, and I agree with them that it was the right one.