This thread explains this phenomenon pretty well and further explains what I suspected with no research done.
I just had no idea that a 8-10Th/s swing was possible at these levels. I knew that charts were screwy but did not dig deep enough to see just how far off the small windows can be. 8-10Th/s still seems to be high but difficulty did just adjust and I am no mathamagician.
I just had no idea that a 8-10Th/s swing was possible at these levels. I knew that charts were screwy but did not dig deep enough to see just how far off the small windows can be. 8-10Th/s still seems to be high but difficulty did just adjust and I am no mathamagician.
mathamagician time - number of blocks per time frame follows a Poisson distribution where variance is equal to the expected number of blocks. So 48 blocks in 8 hours means that you can expect +/- 7 blocks pretty frequently. So +/- 15% is no big deal and extremely common. At 20 TH/s, it is extremely common to see it fluctuate from 17 to 23 TH/s (6 TH/s swing) and common enough to see it fluctuate +/- 30% (from 14 to 26 TH/s).
Note: if total network hashing power decreases rapidly and difficulty remains the same, you'd see an even larger relative variance since you'll have even fewer blocks per time period.
