Lesson Learned: Know your base mechanic's desired result.

[Adam Riemenschneider]

[Lance D. Allen]

This reminds me of the struggle I had before settling on a mechanic for one of my games. It involved a trait you could choose, for which you rolled dice to determine the numeric effect. Originally, it was 1d10 per trait, reroll 1s and 2s. this gave a range, obviously of 3-10, which was acceptable for me.

Others were less inclined to take this trait because of the randomness; You could basically waste one or more of 4 trait picks and get 3 points, and this didn't sit well with some of my playtesters.

So I started fiddling around and discussing the idea. Someone (anonymouse, I believe was the user name he used) suggested something that I eventually adapted and adopted, but not before I futzed with it continually to make sure the desired results were what I wanted.

One thing that really helped me was basic knowledge of how to use Excel to set up formulae with random number generators. I created a single line, and replicated it over about 200-300 rows. I then used another formula to average the results, as well as to allow me if any results were above or below certain maximums and minimums I'd determined. By running the random numbers several times, and watching the average and max/mins, I was able to get an idea for what the results were, and thereby make a judgement.

Eventually, I made one final tweak (involving a maximum number of rolls), but I was satisfied with the result.

Some of my playtesters still weren't satisfied, but more on the principle than because they thought it was unfair or not worth it.

[Adam Riemenschneider]

Glad I'm not the only one.

I got hung up a lot more on the *shape* of the results, instead of the average, min, and max. Here's an example of the things that keep me up at night (the statistical shapes, not the subject matter).

http://www.mwilliams.info/archive/2005/01/men_and_women_are_different_2.php

Scroll down a little bit, until you hit the graph.

Both groups have the same average performance, but the "men" category has a longer head and tail (for lack of knowing better terms) for the curve. The "women" category has a higher peak, and a shorter head and tail.

For some types of rolls, I want a long, flat performance (like rolling percentile, or a single die). Every number has the same chance of being rolled as every other number. And, sometimes I want a higher peak, with more bunching toward the middle; like Gurps use of 3d6. The more dice cast, the more likely that performance of any single roll is going to fall close to the average.

Consider the difference between the three mechanics:
A): 3d6
B): 1d20
C): 2d10

In A, you end up with a total possible range of 3-18. However, odds are really good that result is going to fall between 8 and 12.
In B, you end up with a total possible range of 1-20. Each result is just as likely as any other. To me, this is too unpredictable to use for most types of skill tests, unless the base is quite a large number (such as, Skill Base range of 0-50 + 1d20, vs. difficulties of 10-70). If the Skill Base range is much lower than that, too much of the end result is from this totally random factor. Not great in a Simulationist style of game.
In C, you end up with a total possible range of 2-20. You're going to end up with a bunching result between A and B.

And that's just for the random factor. Then one has to decide on the base total vs. randomized range. Or, ask yourself, "How much of the end effect do I want to be because of Skill, and how much because of random chance?"

Let's use a raw game mechanic of Skill Level + 3d6, vs. a Target Number. With a Skill Level range of 1-5, you can expect to add between 8 and 12 to this number. Because of this, you won't see much difference in how well a Skill Level 1 rolls out, compared to a Skill Level 5. And, if you make Skill Level 5 equivalent to "best in the world" as a skill level, you're going to disappoint the player who invests so heavily in a character trait, only to have flaky results in play.

Go to far in the other direction (say, with Skill Level range of 1-100, +3d6), and there's very little random effect going on. Also, a Skill Level *difference* of 10 or more means it's pretty unlikely that the lower Skill Level will ever outperform the higher Skill Level. Also, a difference of more than 15 becomes impossible to overcome.

So, designers have to ask themselves questions. How "impossible" do I want to make it for an unskilled individual to outperform the most skilled individual? Do I want to make it possible at all? How often do I want said situation to come up in actual gameplay? And so on.

*Then* we can get into the mechanic effects of "special situations." Like a natural 20 always hits! Or rolling a minimum effect on all dice always fails! That sort of thing.

[Lance D. Allen]

Yeah, those unexpected consequences can be a real hassle. In the same game I mentioned above, I had certain stats with beginning values of 1-5, and other stats with randomly rolled values of 1-10. Both sets of stats were essentially rolled for the same types of things... I think the idiocy of this should be readily apparent. I kept this system for quite a while through several playtests, and the longer it went, the more I realized how unworkable it was. But due to the circumstances (the origins of the randomly rolled stats were external and somewhat fleeting) I couldn't find anything that worked that would fulfill my goals.

Eventually, I came up with the system that I'm using now, which involves a roll x amount of dice -vs- y difficulty, where x is equal to the lower based stats, and y is equal to the randomly rolled, more ephemeral stat. I realize that there are those who think that variable number of dice and variable difficulty is a bad design choice, but I typically like the outcomes I've seen with it. I'm not entirely sure how well it'll work for this, as I've not had a great deal of opportunity to test it, but it has been better thought out and is less arbitrary than the last system.

Have you actually plotted curves for your results? I've not really attempted that, so much as I use some tools to lay out some results, and then use a zen approach to determining if I like them.

[Adam Riemenschneider]

I myself haven't plotted out the curves, but one of my playtesters, who is smarter than I am (note: use playtesters who are smarter than you are!), and better at math/statistics than I am (note: see above note), charted out some things and tried to explain it to me. I'm pretty sure I got what he was saying, and was pretty sure it confirmed how I had "Zenned" it out on my own.

The mechanic did some interesting things where I wanted it to, was consistent where I wanted it to be, and was unreliably random where this was appropriate. Made me happy.

Can you run me through a mechanic example for your game? I'm curious to see how things roll out.

-a-

[Lance D. Allen]

In the process of trying to write out an example, I've actually uncovered a gap in the mechanics. Once I get that figured out, I'll be glad to return here and give you an example.

...thanks!