Dominion Strategy Forum

Please login or register.

Login with username, password and session length

Show Posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.

Topics - Thinkaman

Filter to certain boards:

Pages: [1]
1
Forum Games / WitchHunt - Polling for Interest (Mafia, Open)
« on: April 19, 2017, 02:25:00 pm »
Intro - Dominion is probably my favorite "board game", and I was a long-time lurker on f.ds since literally before the Dark Ages. (I think Alchemy had just come out when I started reading?) I don't play much online because I miss isotropic too much, but this place has always had a brilliant community I love reading articles from.  I was teaching some new players the ropes recently, and found myself lost in the wiki and forum posts like the old days.

I felt an urge to contribute myself, but anything clever I have to add about Dominion has already been said better by someone else.  Then I noticed that lots of people had mafia quotes or results in their signatures, and got an idea.

So, I enjoyed a lot of random mafia games growing up, and just shy of 10 years ago I started running a personal variant deemed "WitchHunt." I kept iterating on it and well, here we are.

tl;dr:
  • Fully Open setup
  • 7-25+ players
  • Unique role for each player, bifurcated from team alignment
  • Dead players participate - A collective night action for dead players on each team
  • Strictly two teams - No Kingmaker, Recruitment, ect.
  • No Flip
  • Explicit but Ambiguous Survivals (Last night, theory somehow survived.)
  • Mod-supported/enforced soft-hammers ("deadline votes" standard)

There's a variety of other design goals (improving day 1, universal player agency, clear scum target prioritization, avoiding game end at night rather than lynch) that I could go on and on about, but you get the gist.

I'd copypasta the rules/roles here, but Laura's art is really good so hey: just check out the Official Rules Page with all the pretty pictures.  All the cards + rulebook are there, even an ugly moderator app.


At the risk of making an outrageously arrogant claim, I'm relatively confident that this is the "best" mafia/werewolf/whatever setup. (Shoutouts to The Resistance, which is also excellent but scales better down instead of up.)


IRL game at GDC 2016.

At this point I've probably personally ran about 2000 games; some online, some IRL. (There are no mechanics employed that inhibit either, such as reaction-based decisions for online or private day actions for IRL.) I run about 100 every GenCon in the middle of the deduction alley hallway. It's made to be played a lot. (At the atomic level, much of the design is akin to competitive fighting games.) A number of people in the main STL test group played 500+ IRL games, and a few guys from the old Sirlin.net community played a couple dozen forum-based ones.


These guys. They played... a lot.

We've been through over 360 roles, and let me tell you, there were some bad ones.  I kept a spreadsheet of everything that was tried, so there's a lot of Secret History to be had.


Picture of cumulative prototype decks circa 2015.

At the demands of my friends I ran a successful Kickstarter two years ago, and the game got picked up for publishing by Level99. So now I'll be walking around and see my random mafia setup on the shelf at a game store, or on Amazon or BGG, and it's weird. I never thought I'd make a published "board game", but apparently it just happens. (I could hotlink all those proper nouns, but I don't want to be That Guy.)


Actual Real Objects That Exist™.
See, it's shrink-wrapped, so you know it's good.



Anyway, that's the story.  I haven't hosted a forum-based instance of the game in 2-3 years, because maaan, it's a lot of time to do properly! (I like giving a QT to every player.) When I could go to a convention and run 50 IRL games in less time, the calculus seems wonky, you know?

But I have gotten a lot of personal enjoyment from this community even just from lurking, and you guys seem like one of the more skilled mafia sub-communities I have seen.  So screw it; I'm game to run one for old time's sake, if people are interested.

Cheers!

2
Simulation / Crazy Idea: Genetic Algorithm for Dominion Simulator?
« on: September 13, 2011, 07:07:36 am »
Now that we have super cool simulators by the super cool Geronimoo and now also rspeer, we could potentially get something really crazy going.

Here's a novel flash "game" you might have seen before: http://boxcar2d.com/

It generates random cars and races them.  After so many, it creates a new generation of cars constructed with properties weighted according the the results of the last batch.  In theory, each generation gets better and better, converging to an optimal state.  The algorithm defines how many cars are in each batch, how many and what magnitude of "mutations" occur, and

It's not much of a stretch to see how this applies to Dominion simulation.  Take n strategies, play them all against each other 100/1k/10k times, and then do it again with random mutations of the winners.

I'm pretty sure this could find the optimized Big Money buy pattern consistently--and possibly even find finer grained tuning than we are aware of.  The algorithm would just endlessly add, change, and reorder purchase conditions until it find something better than the old best.  We can further use this to investigate optimal base strategies for any given card (given a defined play logic of course) or combination of cards, such as a full set of 10.

Of course, as the complexity space widens combinatorically by adding more mutation options, it will take longer to converge in any meaningful way. (A 10-card set would take ages to investigate all important combinations.) However, we are talking about a program that could be left to run freely, and one whose work is easily and informally distributable. (The RAM requirements would not scale, and the partial results could be shared just like any other existing simulator templates.)

Now, unless the mutation degree is set excessively high, a genetic algorithm would never make the simultaneous intuitive leaps to find a strategy like Workshop-Gardens from scratch; it would continually find that Workshop is bad incrementally and that Gardens is bad incrementally, and never evolve a good deck with both.  To investigate that sort of strategy, we would have to give it Workshop/Gardens as a starting point.  The key is that the point of this wouldn't be to "solve Dominion", just investigate optimal behavior in given sub-fields we define.

Any interest in this sort of thing?  Especially among super cool people?

Pages: [1]

Page created in 0.126 seconds with 17 queries.