If I had metadata issues in the past I would usually work around them by storing the data set to the repository. Retrieving it would restore the metadata correctly. This would usually happen after an execute R or Python but also transposing a data table. I am finding in 9.21 that this does not work when I run the process. This has nothing to do with the order of the process. To get select attributes for example to see the metadata from a data set retrieved from the repository I have to shutdown and restart Rapidminer and then it works. Running the process with breakpoints sometimes works but when all else fails, restarting Rapidminer is the only option.
On a long process this is a real headache.