I think M/R scripts swallow errors so that it will continue processing the rest of the data. You can find the errors in the Summarize stage