In addition to the timestamp related failures in the test suite, there were some intermittent problems [1] that turned out to be failure of the emacs_deliver_message function. In this series we try to make mail delivery from emacs more robust (mainly replace actual delivery with fcc) [1]: https://buildd.debian.org/fetch.cgi?pkg=notmuch&arch=i386&ver=0.17%7Erc3-1&stamp=1386412394&file=log As I write this, the crypto tests have run successfully for 9155 times. Of course that doesn't really prove anything, but it is twice as many as the previously longest interval before failure.