I've noticed wild swings in the BiB recommendations for fire mages using the adaptive algorithm. Last week it suggested crit was useful for single target. Today it is suggesting to avoid large amounts of crit, and pushing all of my gems to haste:
If I sim my mage using Goroth, I get the advice I had a few days ago: crit remains king for the gear I possess.
I am guessing one of two things have happened. One possibility is the adaptive algorithm was silently "undone" for fire mages because the simulation is transitioning to the newest raid bosses, and not all gear combinations are yet being considered pending further sim data from the global net. If that is the case, then my only suggestion would be putting a warning banner when adaptive algorithms are currently being worked on and therefore subject to higher-than-normal volatility.
Another possibility is that this is just because of natural randomness in the simulation data. To illustrate, I did single sims of my gear with crit gems/enchants vs haste; using different error targets:
- 0.05% error (about 0.4K): 750.7K for crit and 749.4K for haste
- 0.10% error (about 0.8K): 751.2K for crit and 750.1K for haste
- 0.50% error (about 3.7K): 751.5K for crit and 746.6K for haste
Using the 0.05% error target outcomes, the difference between gemming crit vs. haste is about 1.3K or more than 3 times the individual error target for each single simulation. But the same outcomes would be insignificant if I had chosen to sim with 0.50% error target because the difference is less than half the error of each individual sim.
This leads me to wonder how much precision goes into the adaptive algorithm? Even with the glo-net, I imagine there's been a lot of reduction to handle the massive amount of combinations for gear and talents. Could it be that too many things are reduced, or the number crunching precision isn't quite high enough?
The good news is that this hardly matters. My own sims show that going all crit or all haste makes hardly any difference. Maybe even this begs the question if BiB is wasting my time for gemming suggestions since getting it wrong is not going to matter more than a fraction of a percent. But in the spirit of ensuring the best possible recommendations are given, I feel it prudent to speak up since giving the "wrong" recommendation (even if by fractions of a percent) will surely generate controversy by those skeptical of AMRs new approach to gearing.