Just exactly which model did you have, the LIGHTNING-1000 or RBR1000-M?
Perhaps the white arc you saw was just a
lightning bolt and everything is in order?
The RBR1000-M is a multi-rail design, you might have botched the load topology - then again, if the protective circuitry worked as advertised the PSU should have shut itself down to prevent a failure.
1005W at the wall translates to ±820W on the DC side of the PSU so you should have been reasonably safe.
You definitely didn't overload the PSU unless the CPU got highly loaded when you were gone and required much more power. Frankly speaking, Windows is perfectly capable of doing that.
You may have gotten a flawed PSU this time, do remember that the 750W model worked fine even though it was slightly overloaded (with 950W on the AC side, the estimated DC load is again 800-ish).
Truth be told, you're pushing those 5970s
very hard. When you overvolt a card, the power consumption rises proportionally to delta voltage
squared.
I don't think I've seen any 5970 draw more than 280W... ask
DAT, he has a small farm of 5970 cards.
Make sure you enabled all CPU power savings options in the BIOS (AMD cool'n'quiet and C1E) and that you don't use a flawed driver version resulting in totally unnecessary high CPU load when mining.
Vmarchuk is correct, you don't usually want to push the PSU further than 80..90% of its rated maximum continuous load but keep in mind that the actual PSU load should be measured on the DC side. When the best you can do is measure at the wall you need to multiply the readout by the efficiency factor∈(0, 1] which should be at least 0.82 for a bronze-certified PSU.
Pay no heed to that online power calculator, mining is a totally different animal than gaming or any other typical PC use pattern. Trust only your kill-a-watt.
When in doubt choose
Seasonic PSUs - their 7 year warranty period is a fine testament to high quality design and manufacturing.
A manufacturer can't cut corners and expect their product to work flawlessly for seven years.