The estimated next difficulty in my app projects what the next change will be IF the blocks/time solved since last diff change remain constant.
is that in a time of hashrate being added to the network (which has been nearly constant over the last several years) the blocks/time solved since last diff change will decidedly NOT remain constant, but will drop. Perhaps Bitcoinwisdom has over-built this idea into their algorithm, so in times like the present where hashrate seems to be leveling off, it is more inaccurate than during times of rising hashrate. As praeluceo said, any predictor is inherently wrong because the future is unknowable, but I have found Bitcoinwisdom to be a better predictor than all others that I've seen. It certainly converges to the true value as the blocks lead up to the difficulty change.
I'm not talking about using the overall blocks/time to project next change, I'm talking about only the blocks/time since last change.
In other words, we know the target is 6 blocks an hour, or one every 10 minutes. We know how many blocks have been solved since the last change, and we know when the last change was. With those values we can surmise that we're currently solving one every X minutes ... and that value compared to 10 will tell us what the next change will be, assuming we stay at that solve rate.
The downside to this is immediately after the difficulty change the number of blocks solved is so small it makes the estimate vary quite a bit. On the flip side, of course, the closer we get to the next change, the more accurate it is.
That's what I do, and it must be what Bitcoinwisdom is now doing. (I think we're saying the same thing..)
M