I'm playing around with character generation in a purposefully punishing roguelike-inspired D&D homebrew, and trying to evaluate them using Anydice so that each method is roughly comparable. The three methods I'm trying to equivocate are;

  • Roll 4d6 drop highest for each stat, which is real easy to model expected results
  • Roll 15d6 and build stats, so effectively /6 for average ability score
  • Roll 1d20 and 1d4, subtract the lower from the higher.

The last one is giving me some trouble, as it seems simple but I don't know enough script language to write conditional outputs. Would anyone here be able to help?

  • \$\begingroup\$ It seems to me that the average is a poor way of looking at method two. The highly flexible stat building lets you get very high stats where they matter and dump stats you don't care about. Thus, the mean gives a poor representation of the system. \$\endgroup\$ Commented Apr 25, 2017 at 7:54
  • \$\begingroup\$ @JackAidley you are correct that it doesn't really weight the utility of being able to actually reasonably granular scores. I considered dropping it down to 12d6, but that was a little too punishing given that the mean ability scores for 15d6 are already going to be less than the other methods. Given the roguelike/comedy bent of the game, all low or peaks & valleys style scores work well to reinforce wildly chaotic tone, I just want each method to give a roughly the same average result so everybody is screwed over equally. \$\endgroup\$
    – Ambi
    Commented Apr 25, 2017 at 22:54

1 Answer 1


If I'm reading your question correctly, you want to get the distribution of results of the higher of a (d20,d4) pair minus the lower. This anydice program will do it for you:

output [absolute d20 - d4]

It utilizes the built-in absolute value function, which gives the positively-measured distance between two numbers on a number line. (Among many other useful interpretations.)


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .