Babysitter - Shield interaction

[biancospino]

By looking at Natural_Action_Resolution/Normal_Game, it appears that the Babysitter action counts as a killing action

only

when it attempts to kill.

This is problematic, since it means that a Shield would need to wait to resolve until after it can determine whether the Babysitter action is indeed a killing action. Which is extremely bad, as it would create a situation where a

Blocking

action needs to wait to resolve until after some non-manipulative actions have already resolved. This can cause scenarios where the Golden Rule of NAR is entirely sufficient to determine, univocally, a resolution scheme, and yet have said scheme produce an obviously inconsistent result.

Here's the smallest example I could think of, using Normal roles only. Consider the following actors:
V -- Vigilante
B -- combined Babysitter Roleblocker
D -- Doctor
S -- Shield
Now suppose that V target B, D targets B, B targets D, S targets B.
Obviously, D's action is influenced by B's, so B's must happen sooner; and V's action is influenced by D's, so D's must happen before V's. Crucially, B's action is

not

influenced by S's, since it's not a killing action at this stage of resolution and as such it is unaffected by S's shielding; so S can be allowed to resolve later. However, S's action

is

influenced by V's, since it needs to wait to check if any effect would render B's action shieldable, so it must happen at a later time. So, It must be that actions resolve in the order B,D,V,S. So, B babysits and blocks D, which cannot protect B, which causes V to kill B, which causes B's action to become a killing action. So S sees that it can block it, and does it; so in the end, B and S are dead.

But, while this resolution was forced, it has produced an inconsistent result: for now B's action was blocked, so D could protect B, so B should not have died, so B's action should not have been a killing action, so S should not have blocked it.
The important point I want to make is that is not simply a paradox of action resolution, there are oogles of them anyway. The issue is this is a paradox that NAR somehow purports to resolve, but in fact produces a baffling result -- in effect, it pretends that the Babysitter tries to act twice at different points of the action resolution (first to protect&roleblock, and then to kill), when it is actually performing only one action.

I propose two solutions to solve this particular issue:

• change the type of the Babysitter's action so that it is
  always
  both a killing and a protective action. This would make the babysitter almost equivalent, but not quite, to a combined survivalist Doctor gambit Vigilante. (In this case, in the example the resolution would be in the order S,B,D,V, and result in just S dying -- since Shield is worded as "you will block your target's kill actions and you will be killed instead".
  Do note that, if we were using a non-Normal variant of the Shield that redirects the kill action instead, we would still produce a parodox since S and B would block each other, but at least NAR does realize that this is a paradox.
  )
  
• change the Babysitter's action wording as to be just a protective action, and give the Babysitter the passive ability "If you are killed, you will target the target of your Babysitter's ability with a kill. Your Babysitter's ability cannot block this kill.". This would make a Babysitter almost, but not quite, a combined Babysitter-immune Doctor triggered Vigilante. (In this case, action resolution would be as in the example, except it wouldn't be inconsistent, since the Shield would indeed block just the triggered kill and not the rest of the Babysitter action.) This, I believe, is the solution that is going to produce results most akin to what one would naturally expect.

[TemporalLich]

Babysitter is indeed a strange role - it is flavor-wise a protective role but it can act as a killing role

Detective's wiki page seems to suggest Babysitter's kill is indirect (as it only happens if the Babysitter dies that night)

I would be the one to insist that Babysitter only kills when it dies that night however (as Babysitter's wiki page says).

B and D dying would thus be what I think is the correct resolution - Note that in Normal Games there is

always exactly one correct way to resolve minutiae

. The Shield cannot intercept a protection, the Doctor fails to protect the Babysitter due to being roleblocked, the Babysitter dies that night, and thus the Doctor dies as well.

This would be a reason to not have a Babysitter's protection ever be observable as a Killing action. While I would say this leads to the possibility of a strange Observer interaction, the flips would make that evident if you can puzzle out complex Complex Normal setups



and yeah I had confused Shield with Bodyguard multiple times when editing this post, oops.

[biancospino]

While I agree that the Shield should not be able to block the protection, I do not think it can not do that, since there is no mechanism for blocking only part of an action; the only way for the Shield to block the kill is to block the whole package. (In any case, I don't see any world where S would fail to at least soak up B's kill, so it should die instead of D).

That's why my preferred solution would be to word the Babysitter in such a way that its action is just a protection, and should it be killed it passively triggers the generation of

another

action, which is purely a killing one. Which would solve all ambiguities. It would, admittedly, be the only normal thing to trigger action generation, but the alternaive would be to have a normal role that is able to selectively block only part of an action, which to me sounds much less elegant and much less clear.

[DragonEater70]

since there is no mechanism for blocking only part of an action

Herein lies the fallacy.

If a combined Vigilante Fruit Vendor targets player B, and a doctor also targets player B, then only the killing part of the action would be blocked, but not the fruitvending.
Further if a combined Vigilante Fruit Vendor targted a Vigilante-Immune Townie, then even moreso it would become apparent that only part of the action was blocked.

Also, I think the more general fallacy is thinking in terms of "mechanisms", when mafia games are first and foremost adjudicated by their moderator (with interactions between roles usually predetermined in the review phase), rather than by a machine that has to follow a set of rigorously written instructions.

[TemporalLich]

we're talking about Normal Games where the moderator has zero agency over action resolution (moderators have had their games of experience revoked for incorrect action resolution in Normal Games)

[DragonEater70]

I am well aware. I guess I phrased what I wanted to say incorrectly though. I just meant that I don't think the formal existence of a "mechanism" is what defines what can or can't be done in a Normal game, but rather what is established practice. That's due to the concept of Normalness being inherently the concept of what is established practice in mafia games. So I think when dealing with problems of Normality, you should deal with them like in Common Law (i.e. based on precedent) rather than like in Civil Law (based on formulaic rules made in an ivory tower).

[biancospino]

since there is no mechanism for blocking only part of an action

Herein lies the fallacy.

If a combined Vigilante Fruit Vendor targets player B, and a doctor also targets player B, then only the killing part of the action would be blocked, but not the fruitvending.
Further if a combined Vigilante Fruit Vendor targted a Vigilante-Immune Townie, then even moreso it would become apparent that only part of the action was blocked.

That's a well-defined "part" thou. A combined X Y is something that does both X and Y simultaneously on the same target, so it makes sense to roleblock only one of those. I would have no problem if the Babysitter would be called "combined survivalist Doctor gambit Vigilante", but instead it isn't and it is condensed into a single package.

It's entirely possible I may be completely wrong on the issue thou.

Also, I think the more general fallacy is thinking in terms of "mechanisms", when mafia games are first and foremost adjudicated by their moderator (with interactions between roles usually predetermined in the review phase), rather than by a machine that has to follow a set of rigorously written instructions.

See, that's where I fundamentally disagree. I believe that we should strive to word thing in such a way that "resolving actions" be an utterly mechanical task a deterministic machine should be fit to do. Of course, it's impossible to avoid paradoxa (as, say, three Roleblockers blocking each other), both those should be minimized.

[Something_Smart]

In the OP, it seems like you're implying that the Shield blocking the Babysitter kill would also block the Roleblocker action. That doesn't seem to be accurate to me; the Combined wiki page doesn't indicate one way or the other, but it makes no sense to say that a Combined Vigilante Rolecop targeting a Bulletproof would get No Result. (Or that they would fail to kill an Ascetic.)

In my mind, what happens is:
- Doctor gets roleblocked. This roleblock can't be stopped by anything, so it resolves first.
- Doctor fails to protect Babysitter
- Babysitter dies
- Babysitter would kill Doctor, but Shield blocks it. This also causes Shield to die.

There's no paradox here.

[biancospino]

If that's what happens, then yes, I'm satisfied that there is no paradox

[biancospino]

What happens if Vigilante, Shield target Babysitter, and Babysitter, other Vigilante target Townie?

Obviously Babysitter and Shield both die, but does Townie die? I would argue they

shouldn't

, since the Shield is eating the Babysitter kill but not their protection. But for this to happen, we are to assume that the Shield can block only part of the Babysitter action

[DragonEater70]

Well, I've always interpreted the Babysitter's action to either be a protection or a kill, but not both (if it had been both, the Babysitter would, according to strict logic, protect from its own kill). If we apply this to the scenario above, then the resolution would be:
Vig A shoots Babysitter, killing them
Babysitter therefore does not protect Townie A, but attempts to kill them
Shield prevents the kill and dies
As there was no protection on Townie A, Vig B can kill them.

[biancospino]

I always thought it was always a protection, and possibly a kill that specifically pierced its own protection. But I'm starting to feel like I'm probably overcomplicating it.


(I believe the rule as they are currently would produce the resolution in 10, indeed, since I believe the Shield has to block the whole thing.
But it still feels intuitively wrong to me, idk)

[TemporalLich]

In my mind, what happens is:
- Doctor gets roleblocked. This roleblock can't be stopped by anything, so it resolves first.
- Doctor fails to protect Babysitter
- Babysitter dies
- Babysitter would kill Doctor, but Shield blocks it. This also causes Shield to die.

There's no paradox here.

yeah this actually does

seem

right, and also feels right as that uses only NAR's Golden Rule even though Babysitter's action is split into two parts for action resolution purposes (meaning that the combined survivalist doctor gambit vigilante would have the same resolution in this instance at least)

What happens if Vigilante, Shield target Babysitter, and Babysitter, other Vigilante target Townie?

Obviously Babysitter and Shield both die, but does Townie die? I would argue they

shouldn't

, since the Shield is eating the Babysitter kill but not their protection. But for this to happen, we are to assume that the Shield can block only part of the Babysitter action

here's me trying to resolve this strange interaction, considering I've been overruled by a listmod saying that you can shield against a Babysitter kill:

The first vigilante appears to be at the top of the golden rule chain as nothing is interfering with it (this is counter-intuitive and I had to look carefully to discern that fact - it would be intuitive to say the Shield resolves first but this is incorrect)

This means the Babysitter dies that night and thus the Babysitter will kill and not protect

Since you can shield against a Babysitter kill, the Shield will die that night

The Townie dies that Night due to being unprotected and having a kill target them (a Babysitter can't both protect and kill, this isn't quantum physics)

pedit: intuitively speaking, a Babysitter either protects or kills

[biancospino]

(a Babysitter can't both protect and kill, this isn't quantum physics)

pedit: intuitively speaking, a Babysitter either protects or kills

See, I feel this is exactly what is tripping me up, since I was assuming the reverse was true (that is, when the Babysitter kills, it does so by doing both effects and having the kill pierce through the protection).
If this is how a Babysitter is supposed to work, then it should be worded in such a way that the "either" is clear, as it is not.

Just spitballing,
"Each Night, you may target a player. Assuming no interference with your action, that player will be protected from one direct kill. If you are killed,

instead

that player will die."
where

instead

is the key word, which isn't in the standard role PM.

Edit: spelling (would --> will)

[Something_Smart]

Based on that example, I do agree that Babysitter needs to be reworded. I'll ping Ausuka about it.

I think TL's interpretation is more natural and should probably become the official one. I do agree that bianco's reading of it is a little more in line with the wording as it currently stands, although I actually think the most literal interpretation is that if a babysitter dies they will attempt to kill their target, but also protect the target from their OWN kill, which is obviously nonsensical.

[DragonEater70]

Basically, agree with the post below me.

[Something_Smart]

"Each Night, you may target a player. Assuming no interference with your action, that player will be protected from one direct kill. If you are killed,

instead

that player would die."

I think part of the issue with Babysitter is that it's not clear if "would die" is something like Bodyguard dying (completely unpreventable) or if it's a standard kill. I like this, but would favor something even more explicit:

Each Night, you may target a player. Assuming no interference with your action, if you survive the night, that player will be protected from one direct kill. However, if you die, you will instead kill your target.

Phrased that way, it does sound tailor-made to cause awful paradoxes. I'm not sure if it can under current Normal guidelines, but I can easily imagine roles that would cause insane nonsense with it. (The simplest thought experiment is a Reverse Babysitter-- kills their target unless the Reverse Babysitter dies. If a Reverse Babysitter and a Babysitter target each other, it would create an antimatter explosion that would destroy the whole game.)

[TemporalLich]

I've attempted to make Sample Role PMs for the Normal Roles, my Sample Role PM for Babysitter is this

Babysitter - Each night, you may protect a player from a single kill if you weren't killed that night, or kill a player if you were killed that night.

As written, this would make a Babysitter's kill blockable by a Shield but would not make it so that a Babysitter can both protect and kill. (the wording of "weren't killed that night" as opposed to "didn't die that night" is to avoid a Weak Babysitter acting as a Sacrifical Vigilante when targeting non-town)

[Something_Smart]

That is a good point, there is a difference between "die" and "are killed". Although I don't like your sample PM's use of the past tense because that seems like it could really confuse people trying to parse it. I also don't like that it sort of implies the person you attempt to protect and the person you attempt to kill don't have to be the same.

I think intuitively, Weak Babysitter shouldn't kill their scum target, so it is probably better to have "if you are killed, you will instead kill your target".

What a headache. This is why I stick to open games

[TemporalLich]

That is a good point, there is a difference between "die" and "are killed". Although I don't like your sample PM's use of the past tense because that seems like it could really confuse people trying to parse it. I also don't like that it sort of implies the person you attempt to protect and the person you attempt to kill don't have to be the same.

I think intuitively, Weak Babysitter shouldn't kill their scum target, so it is probably better to have "if you are killed, you will instead kill your target".

What a headache. This is why I stick to open games

My Sample Role PMs have the design philosophy of being concise and sightreadable - they are kind of a work in progress however as I'm not feeling 100% happy about them

Goals these Role PMs intend to meet:

• Conciseness
  - These Role PMs should be easily and quickly read and understood. This is the primary goal, as players not understanding Role PMs is harmful to game integrity. Conciseness is achieved by avoiding unnecessary words such as reminder text.
  
• Clarity
  - These Role PMs should be able to be understood without confusion. Clarity is achieved with consistent wording.
  
• Modularity
  - These Role PMs should be able to be dropped in to many setups without many changes. Modularity is achieved by anticipating unusual situations.

The alternate wording of "Babysitter - Each night, you may protect a player from a single kill. You will instead kill that player if you were killed that night." is likely better as it is clear a Babysitter isn't a Simultaneous Survivalist Doctor Gambit Vigilante (which is incorrect)

I'd be willing to make a thread if you want to critique my Sample Role PMs but it would likely be easier if I asked on the MafiaScum discord server

[Gamma Emerald]

I think that wording isn’t very precise either. I could see it being misconstrued as “instead of dying, the player you target is killed in your place” which seems to be the exact error you’re trying to avoid

[biancospino]

Not actually the error we're trying to avoid, but yes, a new plausible error the new wording would produce.
What if we were even more explicit,
"Babysitter - Each night, you may protect a player from a single kill. If you are killed that Night, you will attempt to kill your target

instead of protecting them

."

[Gamma Emerald]

That definitely communicates the practical effects of the role.

[TemporalLich]

Not actually the error we're trying to avoid, but yes, a new plausible error the new wording would produce.
What if we were even more explicit,
"Babysitter - Each night, you may protect a player from a single kill. If you are killed that Night, you will attempt to kill your target

instead of protecting them

."

a bit too verbose for my liking

ability text needs to be concise and elegant more than anything imo

edit:

babysitter is a role that does X if condition Z doesn't happen, but does Y if condition Z happens, so it's quite hard to write elegant Role PMs for

anyway yeah it seems this thread established these minutiae for Babysitter:

- The Babysitter kills if it gets killed that night (dying to Weak is not sufficient)
- The Babysitter's kill can be interacted with as if it were a kill
- Babysitter doesn't investigate as a killing role, but the Babysitter's killing action investigates as a killing action

[biancospino]

"Each Night, you may target a player. Assuming no interference with your action, unless you are killed this Night, your target will be protected from one kill this Night, else you will kill your target."

Unless...else is awkward thou. It would sound better by reversiving it ("if you are killed, you will kill your target, else your target will be protected") but that would give the impression that the BS should investigate as a killing role, which we don't want.