Indeed. It is normal that implementation will find a bug in the spec. Anecdotally, have never found a flawless RFC. Every real-world implementation of an RFC I've ever done has found some discrepancy when it came to compatibility testing.

Yes. But for critical software systems, what excuse does one have not to prove that the software does not match the spec. There is no excuse not to test the spec, but what excuse is there not to prove the software?