Think quoting the snip text from knreed for a 4th time mentioning 5 consecutive valid to be allowed to do the initial processing on ZR alone would be insulting [Still subject to validation definition compliance and not meeting these, going the Inconclusive path, Whatever is assumed to have been forgotten, ''reliable'' was spelled out a device quota having to be maximum [120 per core in a device i.e. no recent direct errors], and turnaround time being avg <= 48 hours. These rules are uniform is my interpretation, no per-plan app.

As I know from some offline comms, not all was taken aboard of server 700 [purposefully] to include the <dont_use_dcf> instruction to clients that would understand this. There was a long long lead time to where WCG is today after the server 700 roll out, which is as I understand, yesterday completing the last step of a series of upgrades to be stable and ready to do further rapid growth, without the accompanying pains. The HCMD2 shorties helped by accident to test for volumetric loads that GPU crunching will add.

--//--

Edit: That old ''reliable'' was after 77 valid tasks [as per that table you don't like or seem to think is wrong] to reach from the start-off assumed 10% error state to get to 0.2% with a limit of 0.1%. No device could go better. That also required a consistent return time of 48 hours on average as with the new rule.