My "non expert" opinion - if the code compiles, it is not wrong. If it does not, then it's worthy of a JIRA ticket to the Apache Hadoop project (maybe one already exists).

Just because Javadoc is missing (somehow), doesn't necessarily mean it's wrong.

In any case, I don't know a single person that explicitly writes mapreduce API code anymore without a higher level abstraction, so what's the use case you're trying to solve?