[JANUARY CHALLENGE]

[callforjudgement]

Key to the Bunker

• 7 Town Bunker Guards
  
• 1 Town Conspirator
  
• 3 Mafia Goons
  
• Day Start; daytalk is not available for either faction
  
• Modified victory condition: town win by eliminating 2 of the 3 Mafia and then going through a Night phase with no town deaths (either by killing the third scum, so that there can be no kill, or by blocking the nightkill)
  
• Each night, a town Bunker Guard may (it isn't compulsive) let a player of their choice into the bunker for that night. (They cannot self-target due to bunker security measures.) If a player outside the bunker submits a nightkill on a player inside the bunker, it will fail. (Membership of the bunker never becomes public until postgame, i.e. scum just have to submit a kill and it will work if no townie targeted the victim or some townie targeted the killer.)
  
• The Conspirator is a modified Neighbouriser; they choose a target during the day phase, and their target is permanently added to their neighbourhood at the start of that Night (thus, it's useful faster than a Neighbouriser normally would be in a non-daytalk setup). Additionally, once during the game, they can attempt to kill someone at night, but it must be a member of their neighbourhood (and if someone lets their target into the bunker, the kill will fail unless someone lets them into the bunker too). Flavour is of someone constructing a team of townies (a conspiracy) to take on the Mafia, but who isn't entirely trustful of all of them.

Bunker Guard is a combination of Doctor and Strengthener. (It started out as just the combination, then I added the flavour, to make it more intuitive how the setup works.) Conspirator is sort-of a combination of Neighbouriser and Vigilante, although it's not a pure combination. So the setup clearly obeys both requirements (a clear combination role and 0 VTs), and also obeys both requirements a second/alternative time in a more twisted way (a modified combination role and a modified Neighbouriser).

The basic idea here is that townies can control the game by only targeting other townies (and lots of them). I suspect there are no breaking strategies simply because scum who know that they've been strengthened can kill anyone they like without giving away information, so town can't coordinate them except via the neighbourhood (and thus probably want to keep scum outside the neighbourhood, especially because it would let scum know who the confirmable townie is). The Conspirator was originally just added in an attempt to shift the balance, but I really like where the role ended up (e.g. they can recruit someone, and kill them the same night if they decide after overNight questioning that their target was scum). The modified win condition helps prevent the game dragging out if a townbloc is correctly identified, and also adds a "semi-White-Flag" effect that helps to balance the setup.

[callforjudgement]

There is a clear strategy... the Conspirator claims Day 1.

A missed kill is only an auto-loss for scum when there's only one scum left (scum can miss a kill at 4:3 without anything happening). As such, I don't see how this plan has a high enough town win chance to be worth trying.

You end up with a random townie dying N1, and a player of town's choice also dying N1; if town miss with both their D1 lynch and N1 vig, which is entirely possible, that takes the game down to 5:3 with one townie confirmed. If they hit with one, the game is 6:2 with one townie confirmed. Town's odds of winning from the first setup are near zero; from the second the best strategy is to forming a three-player townbloc, as you suggest (which is hard in a 6:2, there are 56 possible 3-player blocs and only 20 of them consist of three townies).

In order to win this, town pretty much have to try to block the scum kill early. Claiming would defeat the purpose of that.

[callforjudgement]

No, you are missing something. Two things: 1. Let's assume mislynch and miskill - you are at 5:3 correct... then mislynch would send the game to night 4:3. If the three selected people are townies(who circle-invite themselves to the bunker ONLY if it is a mislynch) then town will survive the night and the three Mafia get voted out(and they have to be Mafia because Mafia would have otherwise killed someone, and won)

a) Finding a three-player townbloc in a 4:3 setup is very hard.
b) It doesn't even help. Scum will just kill the uninvited townie, leading to a 3:3 setup which is a ascum win.

If it is a Mafia lynch Day 2 after mislynch and miskill then the bloc does not activate. You form a bloc of 2. Then you lynch. A Town lynch activates the bloc of 2 who protect each other, if they aren't both townies Mafia wins, otherwise mafia is outed.

So at this point, we have three dead townies and one dead scum going into Night 2: that's 5:2. If scum kills someone (which they will do if you persist with your strategy of "everyone protect the Conspirator"), we're at 4:2 going into Day 3. Assuming a town lynch (as you are above), we'll be at 4:2 or 3:2 going into Night 3. In either of these setups, forming a bloc of 2 (which presumably cross-protect, with nobody else protecting for fear of strengthening scum) is not enough to out scum, they'll just kill outside the bloc (reaching a lylo 3:2 or a winning 2:2 as appropriate).

Actually, I think I see the flaw in your reasoning: in the 3:2 ending, you seem to be assuming that the Conspirator nominates two townies to cross-protect, and

also

that the Conspirator themself is protected. But town only have two protections, and three players to use them on (the Conspirator isn't a Bunker Guard).

A Mafia lynch sends the game to 4:1. A death sends the game to 3:1. Now the Town wins when there is no kill, so the Conspirator now forms a bloc of 2 cross-protecting each other and then No Lynch happens.
If the bloc contains mafia, it's 1v1 at LYLO next day. If not, town wins.

This is an example of the flaw in your reasoning. At 3:1, you have Conspirator, 2 Bunker Guards, 1 Goon. If the plan is for two players to cross-protect, scum will simply kill the Conspirator (who can't cross-protect due to not being able to protect). That takes the setup to a 2:1 where you've gained no information about whether or not scum was in the townbloc.

Note-this is the optimal strategy, but that doesn't mean that it has a very good win rate. It's just not very interesting to play that way.

I'm pretty sure that the optimal strategy is to try to block the scum from killing, thus giving you more time to lynch them (as opposed to your strategy, which gives the scum a kill basically every night). Even if you don't stop the kill, you can probably at least guide it away from the best scumhunters and most townish-looking players, forcing scum to make a suboptimal kill.

[callforjudgement]

Speaking of setups, raw as hell but:

2 Mafia Goon
1 Mafia Neighborizer/Poisoner

1 Weak Neighborizer
7 VT

Poisoner poisons everyone he neighborizes, they die at the beginning of the next night.

Probably needs EV tweaks, but I like the complexity strategy wise.

Isn't it better for the Poisoner simply not to act (except in an emergency)? Doing so would pretty much out them as scum, as it can't be hard to tell which Neighbourizer is which. (Strange though it looks for scum to have an extra kill and not use it.)

I don't think a Weak Neighbourizer is enough by itself to balance an 8:3 (although in an Open, there's always the risk of some sort of absurd claiming strategy). Note that scum will, if run up, claim to have been Neighbourized; any attempt by town to counterclaim this will lead to scum learning the identity of one of the confirmed townies (and if it happens D2, town

can't

counterclaim because they don't know whether the scum Neighbourizer acted, and thus no townie has enough information to know that the claim is false). If the scum Neighbourizer is forced to claim and survives the lynch for that night, they can poison someone overnight in the knowledge that they're doomed the next day anyway, so they may as well take someone with them.

[callforjudgement]

Panic Room

11 players:

1 Mafia "Bomb" (Compulsive Multi-Vengeful)

2 Mafia Goons

1 Town "Recruiter" (Compulsive Day Neighborizer)

7 Vanilla Townies

• Upon the death of the Mafia Bomb, everyone not in the Neighborhood is killed. After this point, the game becomes nightless.
  
• The ability to recruit to the neighborhood is always passed on to the most recent recruit.
  
• If everyone in the neighborhood has died, the game continues on as Vanilla.

I feel like there's the start of a good idea in here, but it might need some tweaking on the exact numbers. I also feel like the EV could probably actually be calculated for this? No clue how to do it, though.

Unfortunately for EV calculations, the setup contains WIFOM (specifically, if scum are recruited to the neighbourhood while the original recruiter is still alve, do they kill the original recruiter (who's confirmable as town) or not?). The existence of WIFOM in a setup makes EV calculations incredibly difficult because they require knowing the optimal breaking strategy, which requires knowing the optimal way to handle the WIFOM (which nearly always involves randomizing your action, but calculating the probabilities can be hard).

[callforjudgement]

Friends Forever

Each player that signs up for the game signs up with a partner.

6* Town Mason Lovers (paired with their partner)
2* Mafia Lovers (paired with their partner)

So essentially 8p lovers but you can choose someone to be on your team no matter what!

That's not an 11p.

Additionally, I think it's equivalent to a 3:1 lylo setup using four hydras, which would be "not mafia" as you don't have any associative tells (everyone knows that players in a pair will be defending each other, and no player has any incentive to defend anyone from a different pair other than reads).

[callforjudgement]

(also it's not lylo, the lovers were meant to be separate)

If town mislynch D1, that's two dead townies. Scum shoot N1 and that's another two dead townies, so it's 2:2 going into D2 = scum win. Thus town have no mislynches, making it a lylo setup by definition.

[callforjudgement]

Adding power roles to Opens is always awkward because they can confirm themself via claiming.

If not for that effect, though, I'd suggest a protective role that only works on VTs (Simple Doctor). This wouldn't work against poison kills.

[callforjudgement]

Ye sorry
Very busy IRL
tw made a Google poll which was nice of him: https://goo.gl/forms/DLt0aUerS8DUs1sB2

Voting starts now and goes until 72 hours from now!

The voting mechanism here is broken in several different ways (read the section I linked onwards in the Wikipedia page).

With the particular variant you're using, if I have three setups I like and one that I kind-of like, my only possible rating choices are 10, 9, 8, 0 and 10, 9, 8, 7. Both of those would be a big distortion. The voting mechanism therefore encourages giving points to setups you think have no chance of winning in order to allow you to give fewer points to others (e.g. giving 7 points to a setup you're sure will lose so that your fourth choice can gain 6 rather than 7, or voting for your own setup to use up a voting slot and thus giving fewer points to other setups); this encouragement has backfired in practice in the past and caused terrible candidates to win. It also makes it more likely that a setup will win if it's similar to several other submitted setups, regardless of its actual merits.

If you're willing to restart the votes, I'd recommend something simple like "rate each setup from 0-10, you can give the same rating to multiple setups if you wish, please rate your own setup as 10 (unless you don't like it) so that all the self-votes cancel each other out, highest average rating wins; using the entire range of ratings is recommended unless you have no strong preference". If not, we should move to something like that for next time.

Incidentally, I had Internet issues accessing the voting pages, so it's possible that my votes have been submitted twice. In such a case, either they'll be identical or one of the copies will be blank, so it should be obvious what my intended votes were.

damn you all for making such awesome setups this is a difficult vote

Could you tone back / retract the language there, please? I know you're joking, but those are strong words to aim at people who are only trying to help.

[callforjudgement]

Yup that's the crux of it.

A score of "1" on the vote card means you give the setup 12 points. A vote of "10" means you give it 1 point.

Sorry for the confusion; once we're through this round it'll be easier to all collab on an easier system I think.

Oh, in that case I also voted in reverse (giving my favourite 10, my less favourite that I liked 9, and so on, with setups I dislike being given -).

[callforjudgement]

It's fine for you to correct the answers. (My votes aren't ambiguous anyway because I rated less than ten setups.)

[callforjudgement]

Private voting is more resistant to people voting tactically, but I don't expect much tactical voting to actually be going on.

[callforjudgement]

I've been trying to design a setup like that for a while, but the problem is that it needs to get really convoluted to prevent the scum simply circle-voting and becoming impossible to lynch as a result. You can do it, but the resulting ruleset looks somewhat arbitrary, and then you wonder why you weren't just playing Vote For Town.

[callforjudgement]

I think Guilt Neighbourhood is pretty scumsided. If not for that, I'd have ranked it more highly. (That said, 2:7 is winnable in practice, and I have a suspicion that 2:9 is actually less scumsided than 2:11, so it may be the best numbers for a vanilla setup with a regular day/night cycle and regular win condition.)

Ignoring my own setup (because I can't evaluate it objectively), NSG's is the most interesting design; my main reservation is that I'm not sure whether it's balanced or not (it may be, it's hard to evaluate).

As for the voting system, I'd recommend something that a) is easy to understand, and b) never causes dishonest rankings to give worse results for the voter than honest rankings (people tend to feel shortchanged if they vote honestly, but the voting system requires a less honest ranking in order for the votes to actually have the desired effect). AV/IRV has the issue that it's fairly complex to understand how the results are calculated (although actually casting the votes is fairly easy); you can also sometimes get an advantage from voting dishonestly, but that's hardly likely to be an issue in practice (as tactical votes in IRV normally require probabilistic coordination among a large number of people, and it's rare for people to regret not having done that).

I still think "each voter rates each setup, highest average rating wins" is likely to be an excellent choice for voting (being easy to understand, both in terms of casting votes and counting them); the main problem is that it tends to miss fine distinctions in which setup is better when two setups are widely liked (voters will often max-rate several options to give a bigger contrast to the setups they dislike, meaning that widely popular options will often get close to the maximum score), but I'm not convinced that that's really a problem; if we end up in a tie for first we can always have a runoff, and it's likely that we'll have discovered multiple good setups in the process. Note that all voting systems are vulnerable to tactical voting, but tactical voting in this system ("range voting") consists of changing ratings while leaving setups in their honest order; there's never an incentive to rank setups out of order.