Wednesday, July 7, 2010

It Doesn't Always Work

Read the blog from the bottom up or it won't make sense.

I didn't have a lot of time to work on JAGS today but what I did do hasn't panned out so well. I'm trying a variant of "Jaws" (teeth) that does more damage once clamped down than the bite-and-release does. I'm calling it Saw-Toothed.

The problem is that the attack-mode has two damages and trying to get the numbers right and have them stay constant or semi-constant throughout the progression (16 AP, 32 AP, 64 AP) is causing me problems. I'll add to that the fact that the attacker's decision to bite-and-hold is something I have to set: in JAGS the amount that an attack "hit by" is important and for a PEN attack a hit by 4 or more is considered a hit to a 'vital' location doing more damage than a hit by 0-3.

Currently I have the simulator set to grab-and-worry on any hit by 4 or better (this stops the attacker from "keeping" a very bad hit). However, in the case of Saw-Toothed it might pay off for the attacker to keep any hit they get (after the grab the target can't block and the strikes don't miss). I haven't tested that yet. It's time consuming to set up the test-run and I'm having enough trouble getting a ratio between bite and bite-and-hold damage to hold steady.

So far it looks like 1:4 is maybe reasonable. That is, if the initial bite does 12 then the worry attack does 48. This is pretty darn high, however so I need to see if the results I'm getting are good. Most recently an 18 PEN 1x per Round Bite with a 72 PEN worry (2x per Round) won 45% against the 16 AP Herd. For comparison, a 1x Attack Bite that scored 50% did 24 PEN (Bite & Worry are the same damage). This means that a 6 PEN deficit would, by the simulator, give us a  48 PEN increase. That seems awfully high to me.

The letters DQI stand for "Data Quality Issue" and it's a generic term for concerns about any result that seems out of whack. When I get a weird result (i.e. I reduce the damage and the % of wins goes up--or I increase the damage and the % of wins doesn't change--or whatever) then I have to immediately look for DQI's.

This is things such as "did I update the Herd Spreadsheet correctly?" It can also be things like "Did I render this attack properly in the Attacks Tab?" On the other hand, the worst case scenario could be "We have a bug in the simulator." It's happened before. A bug now could invalidate all the testing to date if it were pervasive and bad enough. The good news is: all of this has been looked at at least a little. The bad news is: none of the minor bits of the simulator (such as the bite-and-worry code) have been looked at hard enough. A single run (one battle against another) creates hundreds of lines of output when DEBUG is set to maximum level. Our trial runs are groups of 5000 combats so if something weird only happens once in a while we have to determine where by reading the code--not by watching it happen--and then put in some breakpoints or extra debug or something.

This means that if what's going on with Teeth happens to be a DQI? It means I could be debugging this for a long time.

Post Script
I found a DQI in the list of attacks that would explain some of the problems. I'm testing it now.


No comments:

Post a Comment