Not sure I understand this -- are you assuming there’s an existing schema, either a TS type or maybe something else like JSON Schema, and you’re trying to ensure a separate Zod schema parses it correctly?

The usual way to use Zod (or Valibot, etc) is to have the Zod schema be the single source of truth; your TS types are derived from it via z.infer<typeof schema>. That way, there’s no need to take anything on trust, it just works.

- parse as their validation mechanism

- compose small units (close to the type system’s own semantics for their runtime equivalents, trivially verifiable), with general composition semantics (also close to the type system), into larger structures which are (almost) tautologically correct.

Obviously there’s always room for some error, but the approach is about as close to “safe” as you can get without a prover. The most common source of error isn’t mistakes in the underlying implementation but mistakes in communicating/comprehending nuances of the semantics (especially where concepts like optionality and strictness meet concepts like unions and intersections, which are also footguns in the type system).