The spec. could specify exactly the rules rustc actually has including iteration limits, this would freeze rustc (and so it would be undesirable) but it is an option.

The thing we're talking about here has changed after Rust 1.0, Rust 1.0 shipped with a rule that said if you match integers you have to provide a default. In lots of code that feels natural. But not everywhere. If you match all possible integers (e.g. for a i8 that's from -128 to 127), the compiler would say "Not good enough" until you write the default match it will never actually need.

That was fixed with feature(exhaustive_integer_patterns) in 2018 or so AIUI.

But you could imagine Rust being standardised with the old behaviour and, even though it's clearly possible to write a compiler which implements feature(exhaustive_integer_patterns) that would then be non-standard because Rust programs lacking the default match for integers are forbidden in the standard.

The type system is also Turing-complete, therefore non-terminating. It would be arbitrarily hard to write a spec in a non-Turing-complete language such that all implementations would admit the same set of programs, or even (responsive to the original claim) reliably only a smaller set.