Talk:Role

The game and the guidebook refer to the class as the role. I suggest we change it accordingly in the articles of this wiki, too. --ZeroOne 23:53, 24 March 2006 (UTC)

Most of these role pages omit any mention of that roles' speciality spell.--PeterGFin 18:28, 28 April 2007 (UTC)

This should be a huge page. It's one of the most important features of NetHack. It should have a discussion on the differences in roles, each roles special spell, and role difficulty should be merged to this article. And please don't say, "Do it yourself, moron!" because I'm exceedingly lazy and then I'll do it, and it will be shoddy work since I barely know anything at all. Grognor 18:06, 8 June 2007 (UTC)

slashem role table is flat wrong

The table of permissible role/race/alignment combinations seems to be flat wrong. Starting with the first line, the table asserts human,elf,dwarf,gnome,doppelganger, while the code says MH_HUMAN|MH_ELF|MH_DWARF|MH_GNOME|MH_HOBBIT|MH_VAMPIRE. I'm not changing things because I can't wiztest. --Tjr 07:23, 19 August 2011 (UTC)

You don't need to wiztest. Just start a char and see what races you can start as --Elronnd (talk) 05:41, 10 April 2015 (UTC)Elronnd

Attributes table

I have re-done the attributes table. The original table of medians was by User:Derekt75 in Special:Diff/72680, but some of the values seem strange, such as Rangers having median 10 Int and Wis. Not having Derekt75's simulation code, I wrote my own simulation program, but mine calculates the mean and standard deviation instead, which I believe are more useful than only the median. For verifiability, I am including my code below. I used n=10000000 samples for my simulations. Cathartes (talk) 05:54, 6 May 2022 (UTC)

Initial attribute simulation code

I put the code in a github repository, since that's a better place for it than a wiki talk page: https://github.com/vultur-cadens/nethack-analysis/tree/main/attrsim Cathartes (talk) 06:59, 17 July 2025 (UTC)

Sampling attributes from the game

I've done it in a different way for 5.0.0 by enabling the new reroll feature and sampling the attribute values from rerolling 100000 times for each race/role combination: User:Prowler/Attribute_distribution Prowler (talk) 23:44, 26 May 2026 (UTC)

upd: I've run the simulation program for 3.7 and all the results were within ±0.01 of mine. The idea was to get more precise values by sampling from the game itself because, according to folks in IRC, Knights were supposed to have higher Str values than simulated ones because of adjustment due to being encumbered from their starting inventory. Turned out it's not really the case, and after looking more closely at the numbers it would seem that for that to kick in, a Knight has to roll the lowest possible Str and Con and be generated with at least 35 apples and carrots in total at the same time, which basically never happens. Oh well, at least we now have ground truth to validate the simulated results. I've updated the page with simulated output over 10000000 samples. Prowler (talk) 12:15, 27 May 2026 (UTC)

Thank you both so much for your contributions - I'll definitely update those sections for individual roles accordingly when the process of bringing them in line with 5.0.0, though with how much's changed and how much more needs to be accounted for it'll be a while (probably not before June starts, sadly). --Umbire the Phantom (talk) 09:07, 28 May 2026 (UTC)