# Reactor AT DAWN

Reactor AT DAWN is an experimental variation of [Reactor](https://hanabi.wiki/conventions/reactor-intro) which trades clues which get play + discard for clues which get play + info. The idea is that if Cathy is going to be locked, it's very good for her to have information on her cards.

It's called Reactor AT DAWN because the play + info clues are a lot like [Dawn clues](https://docs.google.com/document/d/1dLGcc1iZ_NrxC2rcHDlM6pH_Qna_FrIF8zXMJk2nZGw/edit?tab=t.0#heading=h.5irnh6wujdqv), and I want to pay homage to the conventions that have recognized the value of information this entire time.

## Getting Started

If you would like to get a picture of what the conventions look like before reading this whole document, feel free to skip ahead to [the annotated example games](#Appendix-Annotated-Example-Games).
If you prefer to learn from other humans or have any questions, please don't hesitate to reach out in the [Hanabi Central discord server](https://discord.gg/FJP42bjXf7).

## Motivating Principles

- Action signals: Early on in the game, we want our clues to tell as many players as possible about safe actions in their hand because this allows them to spend their turn drawing cards instead of spending clue tokens.
- Play bias: We want to be giving clues which get plays from as many other players as possible, because playing cards is what unlocks players. To this end, we bias our clue space by trading off some clues which would get discard + play or discard + discard for clues which get play + play or play + play 1-away.
- Slot 1 bias: Newly drawn cards are more likely to be actionable (playable/trash) than already held cards. For this reason, we bias our conventions to make clues that get slot 1 reactions/signals available as much as possible. In particular, our system guarantees that Alice has a clue that gets two plays whenever:
- Bob and Cathy both have a playable slot 1;
- Bob's slot 1 is playable and Cathy's only playable card is already-clued;
- Cathy has no playable cards, but Bob's slot 1 plays into Cathy's slot 1; or
- Cathy has no playable cards, but Bob's slot 1 plays into an already clued card in Cathy's hand.
- Information for locked players: Players who are giving clues benefit much more from information about unactionable cards in their hand than players who can already take actions on other cards. Therefore clues that leave a player locked should give them information about their hand.
- Signal diversity: In a 3-player game, there are 10 actions a player can take from their hand (play/discard slot 1-5), and there are at most 10 clues that are available in a given hand (in the best case, all five colors and numbers appear in the hand). Thus if any two clues to Cathy signal the same action for Bob, there is some action for Bob that cannot be signaled by any clue to Cathy. For this reason, we avoid having different clues get the same reaction.

## Basic Clues

In this section, we will only consider clues touching a single card. Clue focus will be elaborated upon in later sections.

### Single-Play Reactive Clues

The following clues to Cathy are reactive and get exactly one play action between Bob and Cathy:
- A color clue to Cathy on a previously-unclued card.
- A color clue to Cathy on a card which is already clued with color.
- A number clue to Cathy on a card which is already clued with number but not color.

If Cathy holds any playable cards (and hasn't already been told to play them), then it is a "discard + play clue". Bob will react to the clue by discarding, and the slot he discards will tell Cathy what card to play:
- If Bob discards slot 1, Cathy will play the touched card.
- As Bob shifts his discard to the right, Cathy shifts her play to the left.

If Cathy does not hold any playable cards, then it is a "play + info clue". Bob will react by playing, and the slot he plays will tell Cathy information about the touched card:
- If Bob plays slot $n$, the touched card is $n$-away-from-playable.
- If Bob plays slot $n$ but there are no $n$-away-from-playable cards in the suit, the touched card is trash.
- Example: The red 1 is already played and Alice gives a play + info clue on a red 1 in Cathy's hand. Bob plays slot 4, and Cathy sees that there are no 4-away-from-playable cards in the red suit (the red 5 is only 3-away), so knows her red card is trash.
- When the touched card is trash, Bob should react by playing the leftmost slot $n$ for which there are no which $n$-away-from-playable cards in the suit.

(Author's note: I've only really thought about how these play + info clues should work when it is a color clue being given. The number reclue case might be confusing and I haven't seen it come up yet.)

### Double-Play Reactive Clues

The following clues to Cathy are reactive and get an even number of plays between Bob and Cathy (typically two plays but sometimes two discards):
- A number clue to Cathy on a previously-unclued card.
- A number clue to Cathy on a card which is already clued with color.
- A color clue to Cathy on a card which is already clued with number but not color.

If Cathy holds any playable cards, then it is a "play + play clue". Bob will react to the clue by playing, and the slot he plays will tell Cathy what card to play:
- If Bob plays slot 1 and the clue to Cathy was number on a previously-unclued card, Cathy plays the previously-unclued card to its left.
- If Bob plays slot 1 and the clue to Cathy was on an already-clued card, Cathy plays it.
- As Bob shifts his play to the right, Cathy shifts her play to the left.

If Cathy holds no playable cards, but does hold some 1-away-from-playable cards, then it is a "finesse" or "play + play 1-away clue". Bob again reacts to the clue by playing, and the slot he plays will tell Cathy what card to play in the exact same way as for "play + play clues".

If Cathy holds no playable or 1-away-from-playable cards, it is a "discard + discard clue". Bob will react to the clue by discarding, and the slot he disards will tell Cathy what card to discard in the exact same way as for "play + play clues".

### Play + CM Clues

If Alice gives a color clue to Bob on a previously-unclued card, it is a "play + CM" clue. Bob will play, and the slot he plays will tell Cathy how many cards to chop move:
- If Bob plays the newly-touched card, Cathy will not chop move at all, so her chop will be on slot 1.
- As Bob shifts his play to the left (skipping already-clued cards), Cathy shifts her chop to the right.
- We skip already-clued cards since Bob can be told to play those cards with stable Fill-In clues already.

### Stable Clues

Other clues to Bob are "stable" and inherit their meaning from Reactor 2.0. Example stable clues include:
- Left-referential discard clues.
- Fill-in clues.
- Playable rank clues.
- Help yourself clues.
- These can conflict with play + CM clues. For the time being we say that Alice can only give a help yourself clue if Bob can rule out the play + CM intepretation because the card it tells him to play was possible to get with a reactive clue.

## Clue Focus & Targeting

### Reactive Clue Targeting

We inherit the targeting priority from Reactor 2.0:
> 1. We don't reaffirm signals to Cathy's hand that she already has been given.
> 2. Plays take priority over playing 1-aways, which take priority over discards.
> 3. Discarding touched trash takes priority over untouched trash.
> 4. For finesse interpretations, Bob should play slot 1 if he can.
> 5. Remaining ties are broken by leftmost card in Cathy's hand.

### Reactive Clue Focus

The focus of a reactive clue is determined according to the following priorities:
1. Prefer focuses with double-play meaning to single-play meaning.
2. Prefer _Retroactive Fill-In Focuses_ (see below).
3. Prefer previously-untouched focuses.
4. Prefer fill-in focuses (only filled-in cards which were not already known trash or fully known count).
5. Prefer leftmost initial target (the "initial target" is the card that a Bob reaction from slot 1 would talk about).

### Retroactive Fill-In Focus

When a clue fills-in a card in Cathy's hand which Alice is allowed to target with a play signal (either because it is Cathy's leftmost playable or because it is 1-away and Cathy has no playables), Bob always reacts by playing slot 1.
After Bob reacts by playing slot 1, Cathy sees that the clue filled-in a card in her hand which is now known playable.
In this situation, Cathy treats the clue as focused on the filled-in card and takes playing it as her only action from the clue.

Example (filled-in card is playable):
- Only red 1 has already been played.
- Cathy: `r5 b2 b5 [r]2 g5`.
- Alice clues 2s to Cathy.
- Bob knows that this is a play + play clue and that he is expected to target the red 2 in Cathy's hand, since it is her leftmost playable card.
- Bob sees that the red 2 is already known playable as a result of this clue, so he can play slot \#1 to enable a _Retroactive Fill-In Focus_.
- Bob plays slot \#1.
- Cathy sees that Bob played slot \#1 and the red 2 is known playable as a result of this clue, so she treats the clue as focused on slot \#4.

Example (filled-in card is 1-away):
- No cards played.
- Cathy: `r5 g2 [r]2 [b]2 g5`.
- Alice clues 2s to Cathy.
- Bob knows that this is a play + play clue and that he is expected to target some 1-away card in Cathy's hand.
- Bob sees that the red 2 and blue 2 are already known playable as a result of this clue, so he can play slot \#1 to enable a _Retroactive Fill-In Focus_.
- Bob plays slot \#1 and it turns out to be a blue 1.
- Cathy sees that Bob played slot \#1 and the blue 2 is known playable as a result of this clue, so she treats the clue as focused on slot \#4.

### Chop Move Targeting

When a Play + CM clue is given, Alice and Bob need to be on the same page about how many chop moves to give Cathy.
We determine this according to the following priorities:
1. We never CM Cathy if she's already loaded.
2. We prefer to put a trash card on chop.
3. We prefer to put a non-critical card on chop.
4. We prefer to give as few CMs as possible.

## Chop

Similar to Reactor 2.0, players don't have chops by default and are locked unless told otherwise until score 10.
The exception is that after a play + CM clue, Cathy has a chop.

When a player has a chop from a play + CM clue, it is possible that the card is useful or even critical, so players are expected to protect their chops.

## 4-player

The basic idea of play + info clues was actually conceived with higher player counts in mind. The idea is that a clue either gives a play to the clue receiver and gives chops to subsequent players, or gives info to the clue receiver and locks them. Exact details of a 4-player extension TBD.

## Variant-Specific Conventions

### Pink Fill-in Clues

See discussion at https://discord.com/channels/923261838836248646/1381068407293607966

### Prism Bluffs

See proposal at https://discord.com/channels/923261838836248646/1381758388429656094.

## Appendix: Reactive Clue Cheatsheet

| Type of Clue | Prior Clue on Focus | Signaled Actions | Initial Target |
| ------------ | ---------------------- | ---------------- | ---------------- |
| Number | Unclued | Double Play | Left-Referential |
| Number | Color | Double Play | Direct |
| Number | Number (but not Color) | Single Play | Direct |
| Color | Number (but not Color) | Double Play | Direct |
| Color | Unclued | Single Play | Direct |
| Color | Color | Single Play | Direct |

## Appendix: Annotated Example Games

- https://hanab.live/replay/1510461, a 3p No Variant game featuring:
- Turn 4: A reverse reactive finesse.
- Turn 12: A play + info clue which looks like an H-Group bluff.
- Turn 18: A play + info clue which looks like an H-Group finesse.
- Turn 26: A play + CM clue which looks like a referential play clue.
- https://hanab.live/replay/1510507, a 3p No Variant game featuring:
- Turn 5: A reverse reactive finesse.
- Turn 14: A bad stable clue which turns into a reactive clue.
- https://hanab.live/replay/1510485, a 3p No Variant game featuring:
- Turn 1: A help yourself clue.
- Turn 11: A reactive finesse.
- Turn 15: A reverse play + info clue which looks like an H-Group reverse finesse.
- Turn 27: A play + info clue which looks like an H-Group bluff.
- Turn 31 and onwards: Lots of stable direct discard clues for endgame stalling.
- https://hanab.live/replay/1510544, a 3p No Variant game featuring:
- Turn 2: Deferring a reactive clue to give a play + info clue which looks like an H-Group finesse.
- Turn 5: A play + info clue on a 4.
- Turn 11: A play + info clue which looks a lot like an H-Group trash bluff.

## Timo's Scratchpad

4p hypo https://hanab.live/shared-replay-json/415afqawwsmtplukqbfs-vkrlfrcuuakjhiiopmbx-ngdecgnvyphxd,03sdbf-alwcadbqarlcasaeajan-ldatauap1cagaiaokdah-axlabv1cbkbmbabzobb9-aAaBxbaCaDkca3a4bbaE-a6tbaJa2,0