Computational Melee

If you are the right vintage and geek persuasion, you will remember the microgame, Melee, by Steve Jackson that later became part of the role playing system, "The Fantasy Trip," which later evolved into GURPS. Melee was a fun game and I remember playing for hours with my friends in my high school days. I have a long time project, always on the back burner, to make a similar game with new features that does not have the flaws of TFT (I am sure it will have new flaws!). I thought it would be a good exercise to check out the old game for some key playability aspects that I would like to retain in "my" game. You can find the rules to the Melee game online here.

I wrote a simulator in MATLAB for Melee. It leaves out a number of things: ranged combat, hand to hand combat, broken and dropped weapons, pole weapons, temporary DX losses due to damage, and initiative rolls (in the simulator, highest DX always goes first). There 42 possible stat and armour combinations, so the complete Melee game under these restrictions could be represented by a 42 by 42 grid. That seemed like too much data, so I went with one choice versus 42. I started with what I remember as my default character: ST 12/ DX 12(9) with broadsword, leather armour and large shield (4 armour total). I then took the character out of 42 that beat this one most often,  and repeated:

1. 12/12(9)/4: win 30%, average game 4.3 turns against the next entry...
2. 14/10/0: 41% (2.8)
3. 13/11/1: 44% (3.1) 
4. 12/12/1: 45% (6.9)
5. 11/13(9)/5: 36% (5.2) 
6. 15/9/0 

Since I did not do the whole grid, I can't say for sure, but at least from these results it seems plausible that the game has the "metagame" property, that is there is no best character that wins against all others, it is all about dynamically picking the best character against the strategy currently prevalent. I remember the game had an exciting feel, which means that around 5 turns is a good length for a game. The closest to "best" character was #3 above, the 13/11 character with no armour, morning star and light shield: it was only beaten by #4 of all the 42 possible melee character options. However, unarmoured characters would suffer when missile weapons are considered. 

I did 10,000 trials to get the numbers reported above, so they are accurate to 1% with 95% confidence. I am not a statistician, so had to remind myself how these estimates go at the Wikipedia page here.