As discussed in
SERVER-29861 there should probably be mention (or link to discussion) of cases where the number of documents written may differ from the number reported successfully written by the command output.
Example: batch sent, connection closed by server - some documents may have been written. Batch sent, writeconcern satisfied, next batch sent but interrupted - now number written is reported to be size of previously successful batch but some additional documents from the next batch could have been written successfully as well. (it's possible that this belongs with driver or shell bulk method documentation since they split larger batches into batches of 1000).