working on encounter prediction software (for gauging balance)

[blackbird77]

So I posted a previous question about determining what an appropriate encounter would be for a party, based on point values. The best answer I got was that you can't judge whether a combat will be balanced by just looking at the points - instead you have to look at the overall picture, considering OCV, DCV, DC, DEF, SPD, etc.

 

I agree on that, but looking at multiple numbers like that can be misleading, at least to me, so I wrote an app to help me with it. I am still fiddling with it so that it can handle more parameters, but for now it has already helped me tweak a particular combat so that I didn't accidentally kill off the PC's in an unwinnable combat against lower-point opponents.

 

Anyhow, just to describe what it is: it allows me to enter the basic combat stats for the heroes and villains in a particular combat (OCV, DCV, DC, DEF, SPD) plus some flags relevant to my particular set of PC's (are they desolid, can they hit desolid, etc.). Then the software runs the combat, running through the different phases and letting each character pick one of the enemies as a target and attack them, (taking into consideration whether they are desolid) and tracking STUN and BODY to see if they get taken out.

 

I have used this now with a combat i planned for the game i ran last night. I have two PC's each built on 250 points, and i set up a combat where they would fight a 200-pt villain and his two 150-point henchmen. Based on the numbers, i felt it would be a good fight, but that the heroes would pretty easily prevail. Instead, when i set up the simulator to run 100 combats, the heroes only won 1% of the time. So i lowered a couple of the key stats of the villains (reducing their defenses, for instance) and i saw the win percentage start to creep up. I kept nerfing the bad guys until the heroes had a win percentage of about 70%, and then i knew that i had a "hard" encounter but one that was basically winnable.

 

Anyhow, up next i plan to add in the differences between killing attacks and normal attacks, the differences between energy and physical attacks, and then start to bring in area effects and mental attacks too. If anyone has suggestions or interest in following my progress on this, just let me know.

 

PS - i am doing this as a .NET web app with a SQL server backend right now, so it is not immediately easily portable for other people to use, but if there is interest, i can see about modifying it into a windows forms app that runs on access or XML so it could be downloadable by others.

[Folded]

Re: working on encounter prediction software (for gauging balance)

 

This might be interesting. But don't forget that Players will often operate differently than you have modeled above. Some groups would concentrate on the agents, first, then take on the big guy. Others would do the reverse, or take a third option. And movement powers can figure heavily into a fight. A flying opponent alters the tactics of bricks, and a teleporter can mess everyone up, speedsters likewise. A high REC can turn a glass-jawed fool into someone who JUST KEEPS GETTING UP. Then you have the problem of both offensive and defensive Maneuvers of various kinds.

 

In short, this is a very complex piece of work you're attempting. I wish you luck, but don't expect a robust tool to result quickly.

[djkester]

Re: working on encounter prediction software (for gauging balance)

 

Good luck with the software. Once you get it up and running to your satisfaction and make it available online I'd like to see it and perhaps use it. I think it'll help me.

[Christopher]

Re: working on encounter prediction software (for gauging balance)

 

I have a question regarding the target picking: Is it randomly selecting targets (excluding those one can not hit)?

 

Because a random targtting (instead of weakest first), could lead to greater groups getting an statisitcal advantage (when you hit 3 henchmen in that number of phases once each, none will go down. If you hit one until K.O., you will have 1 1/2 down instead).

[IKerensky]

Re: working on encounter prediction software (for gauging balance)

 

Count me in as interested in seeing this

 

I agree that random targetting wont provide a solid basis for analysis. Perhaps better add engagement rules (PC concentrate on weakest foe while NPC share they aggro more evenly, or pairing opponnents...).

[Folded]

Re: working on encounter prediction software (for gauging balance)

 

Running scenarios under multiple attack selection options should ultimately yield the best results, since you can't predict what your players will do. You may want to do some research on how the military runs scenario simulation for ideas. I am very interested to see what you come up with.

[blackbird77]

Re: working on encounter prediction software (for gauging balance)

 

Yes, target selection came up very quickly as one of the "how deep do you want to go on this?" sections. Right now, I have it doing the following:

 

If the character has an attack than can hit desolid, then he looks first for any active opponent (STUN > 0 and BODY > 0) to target. If no such target exists, he picks targets normally.

 

Normal target selection means that a character picks one active opponent at random, but then sticks with that target until one of them is down, even if he keeps missing or failing to penetrate defenses.

 

Overall, this works pretty well so far as a general solution, mostly because I can just have it run 1000 simulations, so any inefficiencies in targeting are getting lost in the noise, but I have a lot to do on that one to bring it up to where it could be. My "to do" list on that one function right now includes:

 

//focus on lowest DCV first

//focus on lowest PD first

//focus on lowest STUN first

//focus on lowest BODY first

//focus on lowest rPD first (considering type of attack - killing)

//focus on lowest ED & rED first (considering type of attack - energy)

//focus on targets possessing specific characteristics, or even flag specific characters pre-combat as the primary target

 

but even for those, I have to decide if the characters should act as though they know in advance what their opponents stats are, or if i want to get into tracking what stats can be inferred based on history in a single combat. (In other words, you know you failed to penetrate his defenses when you rolled average damage - do you still want to go after him or do you assume even your good rolls will not penetrate either?)

 

but i realize even with all these variations, that not every combatant, regardless of PC or NPC, will always follow numeric logic in picking their target. In a combat spread across a large battlefield, the characters will tend to just attack whichever opponent they are closest to, and the numbers-only approach would not simulate that without some randomness thrown in as well.

[Christopher]

Re: working on encounter prediction software (for gauging balance)

 

What about effects that affect a defenders DCV? Like Grabbing, Flashing, Entangle, striking from Invisbility?

You might want to make certain your code can handle those additions in the future.