Expanded the table, efficiency still higher under the nb agent. Once the hundred results are reached, will stop adding more and just follow them through to the end, valid/tool late (impossible quorum), to find the validation rate.

Sadly, a new agent version was launched by berkeley with the usual expectation, members trying again and hoping for miracles. The distribution number jumped significantly in the third week of testing, the average rising to 7.35 copies for the last 20 passing though my tablet and 5 of those are in pval/pver state.

Meantime, the battery usage widget showed the bfa boinc manager sucking 3% in background, now did a force stop on it. The nb manager does not make it to the usage list, meaning, there's the possible source of the efficiency difference.

Been 2 weeks. The good news being the strong indication is, when 'disconnect' the native gui, no more signal 11. Sequentially now 12 valid, 1 pending verification. The repair is waiting to be sent, probably with the new 7.32 science app.

The disconnect is in the right hand top 3 dot drop down menu for native boinc. Connecting works the same way. The core agent continues to run, it just does not get interfered by the gui pulling an update and screen refresh, the spinning wheel centre top appearing when that happens, every second, or whatever other interval chosen.

Now exclusively monitoring from boinctasks on a windows machine. Does not seem the remote rpc taxes the android to interfere with the crunching.

One last with version 7.16 task and waiting on 2 validations to get a final stat for 170 plus results. Starting now with monitoring in detail how version 7.32 of fahv will be doing in production. The beta18 had 3.6 copies average to get to quorum, half of what 7.16 required on average. This is based on 38 results processed on the tab. That said, the beta was probably mostly done by the hands-on crunchers. The total production numbers from the project stats will tell us in a few days if things really improved for many and the volunteers returns to give it another try.

The technicians did not share a single word moving the new science version, maybe absence of fanfare not even a single bugle is the wiser approach this time, once burned twice shy, let the statistics tell.

Hear of hoping.

Now ran 303 work units with the latest fahv application version 7.32. This gave an average success rate of 90 percent in absolute terms. On runtime it's better, 94 percent, which is exactly 10 percent better than version 7.16, see chart in earlier post, noting the now 32 in pv states are assumed to be valid as none so far have gone invalid since 7.32 was 'silently' moved to production around september 25.



The wcg android stats overall are disappointing, just 346k credit for oct.13, also because the average copies needed for valid results dropped from 5.75 to 2.13, not 10 percent requiring an extra verification or repair copy, which is not far to what it is on the pc platforms for quorum 2. The tasks are truly short 4-6 hours with a mean of 4.5 hours, so where are you android crunchers, where have you gone? As the chart shows, no server aborts needed, no too late due too many errors, no invalid.

Btw, it looks like rechencraft rulz the rankings on android. They seem to have figured it out.

Hard to believe, but the android contributions shot up to 588k credit for the next day, no particular extra device registrations. Coincidence i'm sure. Let's see if this holds.