It’s closer to a structured IR that uses a stack encoding, not a machine where the stack is the primary state. The absence of real stack operations (dup, swap, etc.) is not accidental — it shows the stack isn’t meant to be observable.

If you build even a tiny real stack CPU (simulator + assembler + traces), the difference becomes obvious very quickly: the stack stops being syntax and starts being semantics.

So the real question isn’t “is Wasm a stack machine?” but “why does it avoid being one?”

My take: because it’s designed for validation and compilation, not execution as a first-class machine model.

And that’s fine — but then we should call it what it is.