For many 'large' code systems, it will be common to have CodeSystem instances that have 'content' values other than 'complete'. At the moment, that means that when validating, any code value that's not found in the list of CodeSystem concepts has to be treated as a warning - as there's no way of evaluating whether the code is valid or not.
However, the reality is that most code systems have 'rules' that govern what valid 'code' values are allowed to be. For example, for the current release of both SNOMED CT and LOINC, you'll never see a LOINC code that looks like 1234567890, nor will you ever see a SNOMED CT code that looks like 1234-5. If we allowed a code system to have a regular expression, if content weren't 'complete, you still wouldn't be able to definitively say that a code was valid, but you'd be able to very easily assert that many codes weren't valid - because the didn't meet the regular expression. I think this would be a valuable thing and would catch a lot of 'simple' errors.
(If we wanted to be fancier, we could also add an extension that allowed identification of how to extract a checksum value and perform a checksum validation test on the code, but I suspect it would be hard to define that in a 'standard' way. Perhaps an embedded CQL expression?)