Sharp-eyed novi users have found a bug with how the program handles RPM's "epoch" tag. The epoch takes precedence over the version when calculating which is the newer of two RPMs. There are a couple of programs (notably, pilot-link in FC5) for which previous versions of novi mistakenly report the older RPM as the newest.
The good news: I'm testing a fix that should be available for download later this week.
The bad news: if a package has the wrong epoch, then the (corrected) novi calcs in the new version will yield the wrong result.
I'm crafting a novi workaround for the latter issue; but it may be too much of a hack to be worth the effort. The real solution would be to alert package maintainers of the problem -- a newer package having a lower epoch -- such that they can correct it on their end.