Mr. Robot has done it! He can guarantee that he found you the single best solution out of billions or even quadrillions of optimization combinations. And he does this in less than 100 milliseconds on average.
This means that Mr. Robot’s optimizations can’t be beat – his solution is the best you can get. Of course, his optimizations are tied to the stat weights and options you have selected, so make sure you’ve picked your preferred stat weight preset or customized them to your liking. And check in on the options menu too, for things like excluding run speed enchants or legendary meta gems.
How to know if your result is guaranteed optimal
After you click optimize, look for the guaranteed optimal message in the stat section. That ensures your result is the absolute best set of gems, enchants and reforges. Now mouseover it to get stats about your specific result, like how many combinations Mr. Robot had to comb through.
How we know the results are the absolute best?
We wrote a new algorithm using a branch and bound method for those of you familiar with that sort of thing. This algorithm gets the absolute, single best solution. We had it in beta for about 2 months, testing and refining. We continued to do this until every single solution using the new algorithm was guaranteed optimal.
For the final check, we tested it against a sample of 200,000 characters loaded on our site by running them through a brute-force method. We have a program that looks at every single combination of gems, enchants and reforges (instead of using an algorithm). We compared that result to the result from the new algorithm. Not a single solution was wrong – every optimization from the new algorithm was in fact the absolute best result. And the optimizations took on average less than 100 ms each.
And then we popped a bottle of champagne 😉
Why Mr. Robot can’t use the 100% accurate algorithm in every single case?
Speed matters. We want users to have a good experience and not have to wait forever for an optimization. Also, we don’t want a few slow optimizations to bog down our servers (and start costing a fortune). So if the optimization takes too long, Mr. Robot defaults to a faster algorithm that allows for estimation. And you will NOT see the green guaranteed optimal message.
But don’t fret! 85% of those estimations are actually the best solution, we just can’t be sure of it at the time of optimization so we don’t guarantee it. During our testing, the solutions that had to fall back to an estimated solution were compared to the brute-force method’s results. 85% of the estimated results turned out to actually be the best solution! For the 15% that were not the best solution, they were off by an average of 0.1%.
A few things that may result in the estimated algorithm:
- Really low level gear (below iLevel 463)
- Inappropriate gear, like Intellect gear for a Feral druid (we default to the estimation on purpose b/c the gear is so mismatched)
- Rune of reorigination – we have to mess with your weights behind the scene to optimize the rune of reorigination, basically making one of your worst secondary stats awesome. Since we are messing with the weights we automatically do not tell you it is guaranteed optimal, even though it is using the new fancy algorithm. It’s a matter of principal.
- “Incorrect” weights – by this I mean doing something like telling Mr. Robot to force the hit cap, and then putting hit as your lowest priority on the stat weights. This tells Mr. Robot 2 contradictory things: I value being hit capped and hit is worth nothing to me.” So, behind the scenes we raise the value of hit to get you to the cap, but this slows things down and can make Mr. Robot default into estimation mode.
Optimize with confidence!
Mr. Robot is a master at optimizing. He lives and breathes optimization logic. So you can be confident that the result given to you is the absolute best possible solution, given the stat weights you’ve selected/entered. If the results aren’t what you’re expecting (say you see Haste gems when you expect Crit or Intellect), just adjust the stat weights and optimize for another perfect result.