I prevent these stalls by way of a little shell script which I keep running in the background. [Periodically check the task states (boinccmd --get_tasks). If there are some with "state: compute error", trigger a project update (boinccmd --project $URL update).] thus, only few error results accumulate, and the boinc client typically does not resort to fetch the master file.

Most of the bad tasks are now gone. The last batch I downloaded was like 90% good.

I concur. Only 11 bad tasks out of 360 today for 3% error rate. Ten of those I retired with "too many errors"

---