kicking the tires on jev (TypeSafe's System One model) with 2048

I handed jev the current 2048 board and gave it the option of up, down, left, right.. with just that it does about as well as making random moves. I didn't have long to spend on it, so I'm likely missing something.

I don't know how much it changes, but the last version likely has a bug: "gained" is a single number, but you can gain multiple new tiles in a single slide. Also, what are the results for more complex strategies? It's hard to tell how well it's doing if we don't know the difference between a simple strategy and a fancy one. Also we don't know the variance/confidence interval.

Edit: started looking into this and found:

• clockwise strategies (and likely others) have close to bimodal distribution so just a mean is not amazing for comparison: https://project2048.readthedocs.io/en/latest/versus/versus.html#random-vs-clockwise
• basic strategies have massive swings until hundreds of examples are run https://project2048.readthedocs.io/en/latest/simulation/clockwise.html

So in short, if you want to know how jev is performing, you should really graph the distributions and run a thousand games instead of 18. You're mostly getting noise at the moment.