But no repair job was started before returning the WUs. Only after reporting them two repair jobs were created for each WU (still waiting to be sent). So the WU is still holding up the sequential steps processing, even more by being considered invalid and restarting it twice...
So what sense does it make to consider a WU as lost (and therefore invalid) before reaching its deadline but to wait sending a repair job until that WU expires?
And all of the 8 succeeding WUs were offline for nearly three days as well before becoming valid.

Some x-linked side info, 95% of the tasks sent out manage to return the full 100K steps in an assignment: https://secure.worldcommunitygrid.org/forums/wcg/viewpostinthread?post=504350

Right now I have 2 tasks running that were buffered > 48 hours ago, now at 11 and 18%, running high priority. By extrapolation, they're not going to make the 70%, as in 12 hours the 24 hour threshold will happen, soft stop expected. Not had one yet, so will watch how it evolves.

I allowed about 15 WUs to upload all of their intermediate upload files first. Then I forced a scheduler request (not asking for new work) to upload all trickle messages. Then the remaining WUs uploaded their intermediate files and were reported. But that made no difference - all WUs (apart from the 8 already started WUs) went invalid.
When I asked for new work some hours later I got that message for each already reported WU (i.e. 42 message lines):

08.10.2015 12:58:22|World Community Grid|[error] handle_trickle_down failed: null pointer

I will see if I can reproduce this behaviour.

So some invalids later...
The local intermediate upload files are changed as soon as a WU finishes. I.e. up to 90% they are once generated (and hopefully uploaded), but remain unchanged. Reaching 100% the last intermediate files and a result file are generated and all still locally available (not uploaded) intermediate files are changed somehow - they get a new timestamp and differ if they are compared binary. Since I do not know their structure I cannot tell the differences but if they are uploaded and the WU is reported it becomes invalid. If I replace the changed upload files with the original versions before uploading, the WU becomes valid, also if there was already an upload for that WU before (well, replacing the files only simulates that situation...). So going online before reaching 100% is sufficient to get a valid result, but going online after the WU finished makes it invalid...
Also, if an intermediate file (e.g. the first one for 10%) and the associated trickle is uploaded the processed time and some claimed credit shows up in the result status page after approx. two minutes. After completion of the WU the same action with the now changed intermediate/trickle file has no influence on the result status page - the values remain still 0.

These are only observations, maybe they have nothing to do at all with the offline problem. Will do more testing if time allows (need 18 hours for the next WU to complete before I can start testing...).

Matthias