# Reactor 2.0 Winstreak Agreements

Current Winstreak: 11
1. https://hanab.live/replay/1569830
2. https://hanab.live/replay/1569857
3. https://hanab.live/replay/1570593
4. https://hanab.live/replay/1570655
5. https://hanab.live/replay/1570898
6. https://hanab.live/replay/1571396
- Desync on turn 12: Timo didn't react to a reactive clue.
7. https://hanab.live/replay/1573458
8. https://hanab.live/replay/1573617
9. https://hanab.live/replay/1575799
10. https://hanab.live/replay/1575835
11. https://hanab.live/replay/1587398

Old Winstreak: 16
1. https://hanab.live/shared-replay/1529187
- Desync on turn 20: Fafrd thought g3 elim would be global and make sjdrodge loaded.
2. https://hanab.live/shared-replay/1529216
3. https://hanab.live/shared-replay/1529606
4. https://hanab.live/shared-replay/1530483
- Desync on turn 11: Fafrd played a r2 which was told to discard
- Desync on turn 11: Doodles thought the filled-in trash p1 was the focus of the reactive clue
5. https://hanab.live/shared-replay/1531274
- Turn 5: Is Bob allowed to gentleman's discard his queued playable without it turning Alice's clue into a reactive clue?
- Possible Principle: After Alice clues Cathy, Bob should only take conventionally expected actions unless he is okay with Cathy trying to interpret it as a reactive clue.
6. https://hanab.live/shared-replay/1535371
7. https://hanab.live/shared-replay/1537460
8. https://hanab.live/shared-replay/1539095
9. https://hanab.live/shared-replay/1539112
10. https://hanab.live/shared-replay/1539137
11. https://hanab.live/shared-replay/1545676
12. https://hanab.live/shared-replay/1547792
13. https://hanab.live/shared-replay/1558980
14. https://hanab.live/shared-replay/1560008
- Desync on turn 21: Fafrd didn't deprioritize the gotten p1 for initial targeting
15. https://hanab.live/shared-replay/1560065
16. https://hanab.live/shared-replay/1566430
17. https://hanab.live/shared-replay/1569814
- playable blindness on turn 12 leads to critical discard

This document is structured into sections discussing conventional design questions and possible answers to them. During play, players can refer to the table below for the answer we are currently rolling with.

## Current Decisions

Alternative table. This is Fafrd's preference for easy reference and as a way to remember recent discussion conclusions.

|Row| Subject | Current Policy | Motivation |
|------|---------|----------------|------------|
|1|Good touch playables|Good touch playable cards are queued plays. If there are too many good touch cards to be all good, none are queued without further signaling, but the rightmost must play before discarding chop.|This is simple and easy to use|
|2|Gentleman's Discards|GD cards are queued plays.|Matches with good touch|
|3|Bad touched cards|If a bad touched card is queued to play, fixing is higher priority as a reactive target than signaling a playable.||
|4|Fill-in Focus| A reactive fill-in clue can focus on a filled in playable which is possible valid reactive target (not gotten). If Ava has no playables, retroactive finesse focus applies. ||
|5|Elim|Elim does not affect clue interpretation.|Better for sync and error recovery|
|6|Error recovery|After an error, any possibly responsible clues are called off. If Ruben goes back for such a clue, Ava is expected to follow.||
|7|Crosshand dup extraction|If Ruben must agree to an apparent bdr as a reactive target, he targets Alice's initial target as dup extraction.|We want to kill crosshand dups to avoid double locks|
|8|Locked clues to Cathy|If Alice is locked and Bob has a known discard, Alice can clue stable to Cathy|This can help Bob unlock Alice|
|9|Known dup targeting|Alice is allowed to target Ava to play a card which is known queued in Alice's hand, though it is the lowest priority play in Ava's hand. It is higher priority than any finesse interpretation. Cards which play through a globally known play in Alice's hand are also the lowest priority finesse targets.||
|10|Negative fill-ins|Stable negative fill-ins are allowed for ranked clues that touch a new card|This allows clueing a 5 to make a 4 play|
|11|Conflicting reactions|If Ruben is asked to react to a new reactive clue while an older one is pending, the old signal is cancelled and Ruben must respond to the new one. If Ruben goes back for the old signal later, Ava is expected to follow.||
|12|Bluffs|Ruben should not entertain possible bluffs|Better for sync|
|13|Shared knowledge|Reactive targeting is based on global knowledge, not the shared knowledge between Alice and Ruben.|Ava's understanding and reads can matter|
|14|Same hand playable dups|If Ava has a clued and an unclued copy of a playable card, the clued copy is the agreed target for reactive play signals.|This prevents bad touch|

## Recent Agreements

### Good Touch Critical Bomb
If a good touch card that has exactly one playable identity bombs and that identity is critical, all other good touched cards queued as exactly that identity are unqueued.

### Expected Deductions
Any touched cards in Ava's hand (including previously touched) that are proven trash directly by the reactive targeting rules + Ruben's reaction (without reference to any previous turns) are marked globally as trash.

Ava is not expected to reason about what reactions may be impossible. The team should mark trash as if any reaction was possible for Reuben.

### Help Yourself Promise Delay
When a help yourself clue touches multiple unclued cards, position is not promised until the focused card is immediately playable.

### Good touch queue timing
A card becomes queued from good touch the moment that it becomes immediately playable or trash from global empathy, or will become playable after that players other queued plays.
If there are more touched candidates than possible playables, none are queued without further help.

### Hot and cold clues

The Cold Phase starts when score hits 10. The card drawn by the 10th play is the first chop.

Alice is cold = Alice is locked or it's the cold phase and Alice does not know about a play

Bob is reversible whenever he has a play. He is also reversible when he has known trash and Alice is cold or Cathy is locked

A player is skippable whenever they have a queued play and the other doesn't. A player is also skippable if they have known trash and Alice is cold

The meaning of "skippable" and reversible is given below in [How do we tell whether a clue is stable or reactive?](#How-do-we-tell-whether-a-clue-is-stable-or-reactive)

> We say a player is "skippable" if Alice is allowed to give a clue which "skips over them" (i.e. gives them an opportunity to react) without assigning them a safe action.
> We say that Bob is "reversible" when Alice can clue Bob, expecting him to take an already known action and entertain a reverse reactive/bad stable clue.
>
> When Bob is skippable, Alice can give good stable clues to Cathy.
> When Bob is not reversible or Bob is skippable and Cathy is as well, Alice can give good stable clues to Bob.

### Stacked reactive clues
When Alice is hot, the first player out of Bob and Cathy to run out of plays is Reuben and the signalled actions are added to the end of Reuben and Avas' queues.
Targeting in Ava's hand is based on play stacks at the time that Reuben runs outs of plays and can interpret the reactive clue.

### Focusing gotten cards
Initial reactive target deprioritizes queued plays.

### Targeting delayed playables
Delayed playable cards can be targeted iff the intervening card(s) are all in Ava's hand

### Reverse reactive cancelling
If Alice clues Bob and he immediately takes a play/discard which was not gotten prior to Alice's clue, the team treats Alice's clue as stable

### Direct crosshand dupe extraction
Contrary to what the wiki says, stable clues touching new cards and filling in a cross-hand dupe are focused on the new card. If the clue is only a reclue, then it can indicate a discard of a crosshand dupe.

### Good touch overload avoidance
Good touch playable cards all of whose playable identities are already fully known elswhere are treated as trash.

### Good touch default actions
If you have more good touch candidates than playable identities, none are automatically queued, but you cannot be given ptd, if left to your own devices, your default action is to play rightmost.

## What is the focus of a reactive clue?

A card touched by a clue is called its "focus" if the clue would give the same signals were it to touch only that card. This section concerns how the focus of a reactive clue is determined.

### Discussion

Why prefer filled-in playable focuses? (A)
- It matches intuition from stable clue focus.
- It means we don't need a special case for how Cathy interprets play slot 1 reactions on clues which fill in a playable.

Why prefer newly clued cards? (A, B)
- It matches pattern recognition from ref-sieve.
- It increases variety of available focuses. For example, when the leftmost red card in hand gets clued with rank, the next leftmost red card becomes focusable with red.
- It means that our initial signals are more often on unknown cards, which is good because we want play slot 1 to be a viable reaction as much as possible. (Although we could again special case play slot 1 reactions.)

Why prefer fill-in focuses over no-info reclue focuses? (A, B, C)
- It matches intuition that a clue which fills in a card is not a "no-info clue".
- It trades single-play clues for double-play clues.

Why prefer focuses which cause the leftmost initial target? (A, B, C)
- It means we can more often target slot 1 with initial play/initial discard signals.
- Since our rank clues are left-referential, it gives them predictable focus availability.

Why trade single-play clues for double-play clues? (B, C)
- There are some game contexts where the former are valuable and the latter are often not enough worth giving. (e.g. occupied clue-giver, endgame)
- It enables more finesses.
- It is more important to have a diversity of available play signals than available discard signals, since there are usually fewer slots in hand that are okay to play signal than slots which are okay to discard signal.

Conversely, why trade double-play clues for single-play clues? (A)
- It guarantees that we have a clue which gives an initial discard signal to slot 1, so we can get play slot 1 + discard slot 1 in either order.

### Current Answer: Stable Intuition

We determine the focus of a reactive clue according to the following algorithm:
1. If the clue fills in a playable card, the leftmost such card is the focus.
- Here "playable" means that the card is a valid play target for this reactive clue.
2. Otherwise, if the clue touches any previously-unclued cards, the focus will be the newly-clued card which results in the leftmost initial target.
3. Otherwise, if the clue fills in any cards, the focus will be leftmost newly-filled-in card.
4. Otherwise, it is a no-info clue, and the focus is the leftmost touched card.

Equivalently, we can describe this in terms of our priorities for the initial target of a reactive clue:
1. We prefer initial targets which the clue filled in as playable.
2. We prefer initial targets which were previously unclued.
3. We prefer initial targets which were previously unknown but got filled-in by the clue.
4. To break ties, we prefer the leftmost initial target.

In addition, the "[retroactive finesse focus](#Retroactive-Finesse-Focus)" convention described below is on.

### Previous Answer: Double-Play Bias

We determine the focus of a reactive clue according to the following priorities:
1. Prefer focuses with double-play meaning.
2. Prefer newly-clued focuses over previously-clued focuses.
3. Prefer focuses with leftmost initial target.

In particular:
- Number clues are focused on filled-in cards before new cards and already number-clued cards.
- Color clues are focused on new and filled-in cards before cards already clued with color but not number.
- Color clues make no distinction between

In addition, the "reactive tempo clue" and "retroactive finesse focus" conventions described below are on.

#### Cheatsheet

In order of focus priority.

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

### Untested Answer: Strong Leftmost Bias

We determine the focus of a reactive clue according to the following priorities:
1. Prefer focuses with double-play meaning.
2. Prefer focuses with leftmost initial target.

In addition, the "reactive tempo clue", "reactive direct play clue", "retroactive finesse focus" conventions described below are on.

### Retroactive Finesse Focus (On)

For Bob (the reacter):
- If a reactive double play clue fills-in a 1-away card in Cathy's hand and Cathy has no playables in her hand, Bob always reacts by playing slot 1.
- If a reactive single play clue fills-in a 1-away card in Cathy's hand and Cathy has no playables or trash in her hand, Bob always reacts by playing slot 1.

For Cathy (the receiver):
- If a reactive clue fills-in a 1-away card in Cathy's hand and Bob reacts by playing the connecting card from slot 1, Cathy treats the clue as focused on the filled-in card and takes no further actions from it.
- If Bob reacts any other way, Cathy determines the focus of the clue normally.

### Reactive Tempo Clue (Off)

For Bob (the reacter):
- If a reactive clue fills-in a playable card in Cathy's hand which is the expected target in her hand (i.e. the leftmost playable), Bob always reactive by playing slot 1.

For Cathy (the receiver):
- If a reactive clue fills-in a playable card in Cathy's hand and Bob reacts by playing slot 1, Cathy treats the clue as focused on the filled-in card and takes no further actions from it.
- If Bob reacts other than playing slot 1, Cathy determines the focus of the clue normally.

### Reactive Direct Play Clue (Off)

For Bob (the reacter):
- If the leftmost newly clued card touched by a reactive color clue is playable and the expected target in her hand (i.e. the leftmost playable), Bob always reacts by playing slot 1.

For Cathy (the receiver):
- If a reactive color clue touches new cards in Cathy's hand but does not reveal any playables and Bob reacts by playing slot 1, Cathy treats the clue as focused on the leftmost newly clued card and plays it.
- If Bob reacts other than playing slot 1, Cathy determines the focus of the clue normally.

## How are reactive targets chosen?

When Alice gives a reactive clue, she and Bob (the reacter) have to agree on what Bob's reaction should be, and therefore what the clue will signal to Cathy. In particular, Alice and Bob look at Cathy's hand and determine a card which is the "expected target" and whether to give it a play or discard signal. This section concerns the algorithm they use to find this expected target.

### Discussion

- How do we handle the multiple copies of a same-hand dupe?
- How do we handle cross-hand dupe extraction clues? (Play + discard clues when the receiver has no trash.) We discussed always play slot 1 if you can, but the doc says to try to target a card which could be a duplicate of a _clued_ card in the reacter's hand.
- Do we want to prioritize discard targets more often? Maybe at low clue counts or with not much known trash in hands?

Faf's discussion notes:

Same hand dups:
Unlued copy is preferred target for discarding, clued copy preferred for play signals. This means the unclued copy will be ignored fully for play signals even if it's the leftmost playable.

cross hand dup extraction:
if alice gives a play + discard clue and cathy has no plays or safe discards, bob prefers the initial target + slot 1 reaction as dup extraction, (and leftmost bdr otherwise?)

global vs mutual targeting:
the target in cathy's hand should only consider globally gotten cards, even if alice and bob could agree that another card was already known playable.

### Doc Answers

The reactor 2.0 doc has multiple sections which answer this

https://hanabi.wiki/conventions/reactor-intro#the-expected-signal-to-cathy

- We don't reaffirm signals to Cathy's hand that she already has been given.
- Plays take priority over playing 1-aways, which take priority over discards.
- Discarding touched trash takes priority over untouched trash.
- For finesses interpretations, Bob should play slot 1 if he can.
- Remaining ties are broken by leftmost card in Cathy's hand.

https://hanabi.wiki/conventions/reactor-intro#the-expected-signal-to-the-receiver1
1. Discard play-queued bomb.
2. Play leftmost playable.
3. Play focused 1-away.
4. Play leftmost 1-away.
5. Discard leftmost unknown touched trash.
6. Discard leftmost trash.
7. Discard the closest card which could be a dupe of a touched card in the receiver's hand.

## How do we recover from errors?

error recovery:
after a bomb bob may give up on the original clue, but if bob chooses to go back and find the intended reaction then cathy is expected to also figure out what happened.

## What do Gentleman's Discards do?

## When do discards give conventional elimination notes?

elim:
any useful card that discards can trigger elim, however this elim does not affect future clue interpretation by default

## How do we tell whether a clue is stable or reactive?

We say a player is "skippable" if Alice is allowed to give a clue which "skips over them" (i.e. gives them an opportunity to react) without assigning them a safe action.
We say that Bob is "reversible" when Alice can clue Bob, expecting him to take an already known action and entertain a reverse reactive/bad stable clue.

When Bob is skippable, Alice can give good stable clues to Cathy.
When Bob is not reversible or Bob is skippable and Cathy is as well, Alice can give good stable clues to Bob.

Current answer: https://hackmd.io/H5OrlGYTT1y_OoBchjuK9w?both#Hot-and-cold-clues

### Discussion

Possible answers

"hot and cold clues" (current) https://hackmd.io/H5OrlGYTT1y_OoBchjuK9w?both#Hot-and-cold-clues:

- skippable = has known play and other player doesn't OR (has loaded action and Alice is cold)
- Bob reversible = has a play OR (has known trash and (Alice is cold or Cathy is locked))

"sodium headcanon":
- skippable = has a known play or (has known discard and Alice is locked)
- Bob reversible = Bob is skippable and Cathy isn't

"Timo Proposal":
- skippable = has a known safe action (play, trash, stable discard target)
- reversible = has a play OR (has known discard and Cathy is locked)

Motivation for reversibility when Bob has known discard and Cathy is locked: Hypothetical reverse reactive green clue on https://hanab.live/replay/1539050#14.

Relevant turn: https://hanab.live/shared-replay/1539095#24

### Possible Principles
Principle: We never skip over locked players. In other words, a player can be skippable only if they have a known action.

Principle: Reactions are expected to occur in the round that their corresponding reactive clue is given. If a reaction does not occur in the same round, it is called off for conventional intepretations. If it is later demonstrated on the
Consequence: When Bob and Cathy both have queued plays and Alice gives a bad stable clue to Cathy, Bob must react immediately rather than waiting in case his reaction plays on top of a queued play.
Consequence: If Alice gives a reactive clue to Cathy and then Bob gives a clue instead of reacting, the reactive clue is called off. (Although Bob can go back and take his reaction to put it back on.)
Consequence: If Alice gives a reversible clue to Bob and then Bob gives a clue which gets an action from Cathy, Alice's reversible clue is called off. (Although Cathy can go back and take her reaction to put it back on.)

Principle: If Bob is given a reversible clue and then cuts off his information channel by giving Cathy an action, then he can never assume the clue was stable.

Principle: Reuben must react to a bad stable clue in the same round it was given if he has the opportunity, otherwise Ava will assume it is good stable.

### Good Stable Clues

Types of stable clues:
- Direct play clues
- Referential discard clues
- Fill-in clues (filling in a previously unknown playable/trash card)
- Playable rank clues
- Fix clues (revealing a queued play was trash)
- Direct discard clues: clues on new cards which give empathy trash
- Dedupe clues: Clues which reveal a cross-hand dupe
- Help-Yourself clues
- Starting hand stalls: rank targeting slot 1 on the first turn of the game
- Rank stall: rank clue on new cards to a loaded player
- Fill-in stall: Clue to a loaded player touching only already-clued cards and giving new information to them

Question: Can we give help yourself clues to Cathy when Bob is loaded?

#### Timo Proposal

The following list gives the priorities which Alice must follow when deciding between stable clues. If Alice gives a stable clue and an intervening player can prove that a lower priority clue was given when a higher priority clue was available, they should react to turn it into a bad stable clue.

1. fix clues, direct play clues, fill-in play clues, and playable rank clues
2. referential discard clues on trash, fill-in clues on trash
3. any clue besides a help-yourself clue
4. help-yourself clue

Question: Should given trash signals be allowed at low clue counts?
Question: Should direct discard clues on cross-hand dupes be allowed over discard clues on trash?

Relevant turn: https://hanab.live/replay/1528214#13
Is this blue clue a good stable clue as a help-yourself? Under this proposal, no. Finesse into y3 would take precedence if there wasn't !y2 context. Is it then a double discard clue under this convention??

#### Other Ideas

We could say that Alice should find a stable clue to Ava which requires nothing from Reuben's hand if possible, e.g. If Reuben has an unknown queued play, Alice should giving a stable clue to Ava on a card Reuben knows he is not playing, and Alice should try not to give a stable clue to Ava on a card that plays on top of his unknown play, since this forces him to decide against a finesse.

## How do we treat good-touch-playable cards?

option: default action is to play good touch card but any signal takes precedence
option(current): good touch action is queu-ed same as a normal play unless forced otherwise

Timo's Good Touch Proposal: https://discord.com/channels/923261838836248646/1349044707661189222/1355776392394178751
> I propose the following configuration (at least for 3 player - I make no comment on 4p+):
> - Players are expected to play from good touch before discarding chop if they are given nothing else to do, but good touch plays do not count as queued actions for the purpose of determining reacter vs receiver.
> - Stable "playable rank" clues (e.g. stable 1s clues) still queue all of the touched cards because the purpose of such a clue is explicitly to give a play signal to the touched cards.
> - It is legal for a good touch playable card to be targeted by a reactive clue, so if the reacter plays a good touch playable card after a clue, then they are demonstrating that as their reaction.
> - Targeting priority for the receiver treats good touch playable cards like any other immediately playable card. (Typically good touch playable cards will be on the right side of the hand though, so will be the last to receive play signals.)
> - Targeting priority for the receiver prioritizes discard signals on cards which are *not* good touch playable. (This is because receiving a discard signal at all indicates that there are no playables, so instantly gives a known trash note to any good touch playable card for free.)
> - Note that finesses still have higher priority than discard signals on cards which are believed to be good touch playable.
> - My rationale for this choice is that a scenario in which Cathy's default action is to bomb trash is not much different from a scenario in which Cathy's default action is a dangerous chop discard. In either case, should Alice give a stable clue to Bob, Bob will typically clue Cathy. In the case where Cathy's chop is dangerous, we already prioritize finesses over trash discard signals, so this seems no different. One benefit of this is that once Cathy plays a 1-away card, she can infer for free that her good touch playable card is trash.

## How are reaction slots determined in the presence of expected/default actions?

https://hackmd.io/-_RWEN-ASTq7sdP9Nujjtw

## Interesting Turns

- https://hanab.live/replay/1523767#17
- Do we target g4 or b3? (Probably just g4...)
- https://hanab.live/replay/1523652#14
- This was a really cool deck where play + discard clues really shined. This turn was a cross-hand dupe extraction, a move Timo had never seen before!
- https://hanab.live/shared-replay/1524020#11
- Is this a bad stable clue? What cards is Alice allowed to target with a stable discard clue in Cathy's hand?
- Timo Proposal: Good stable clues must target trash if possible, then non-criticals if possible.
- https://hanab.live/shared-replay/1524020#13
- If faf gives a bluff/finesse on y4 here, can Timo discard slot 1 as y3 should faf have it?
- https://hanab.live/shared-replay/1524020#23
- Is r3 queued due to elim notes? When do elim notes affect clue interp?
- https://hanab.live/shared-replay/1524020#29
- Does Faf have ptd on slot 1 after the 4 clue? What should this do if the card referred to by the new 4 was not also the chop?
- https://hanab.live/shared-replay/1524045#19
- Is b4 queued?
- Proposal: Stable discard clues from locked players give a "chop" note, not a "kt" note, so discarding it is a default action but e.g. can count as a reaction if preceded by a clue.
- https://hanab.live/shared-replay/1524045#27
- This worked as a reverse reactive despite r3 only being known from elim prior to the clue. Can we adopt some agreement that allows reverse reactives to work even when just Alice and Bob agree that Bob has a loaded action as long as Bob demonstrates the action right away?
- https://hanab.live/replay/1524100#7
- Stephen clues 1s which conventionally should be reverse reactive but maybe just works since Faf plays right away
- https://hanab.live/shared-replay/1528010#24
- Can Libster expect Timo to gentleman's discard p4 and therefore consider Fafrd queued and give stable fill-ins to Timo?
- https://hanab.live/shared-replay/1528172#44
- Is blue here an endgame stall?
- https://hanab.live/shared-replay/1528198#28
- Does Timo have a clue which doesn't force Stephen to draw?
- https://hanab.live/shared-replay/1528198#37
- Bob gets to demonstrate that it's not reverse reactive
- https://hanab.live/shared-replay/1529606#11
- What happens when Bob clues after receiving a reverse reactive clue?
- https://hanab.live/shared-replay/1536717
- Desync on turn 0: Decided to score 0
- https://hanab.live/shared-replay/1539095#40
- Do we target r5 here? What if sodium's r4 was known?
- https://hanab.live/shared-replay/1542437#26
- It seems like a bad convention for timo to have a chop here. What if his slot 1 was y5?

### Challenge Games

- No starting clues:
- https://hanab.live/replay/1536725
- Designated cluegiver:
- https://hanab.live/replay/1538560
- Scarce ones:
- https://hanab.live/replay/1509452
- The steak game:
- https://hanab.live/replay/1545627

## Ideas to revisit

### Short term questions

Should gds of unknown plays promise position?

what does 5s to slot 1+2 mean?

how does focus work on a no info reclue? do we focus leftmost and then decide if it was no info or not, or do we focus the non-filled in card

We have had a few times that we tried to treat signalled discards as deprioritized for new clue focus/targeting, but the current agreement says that is for queued plays only. Should it be expanded to include discards?
-https://hanab.live/shared-replay/1546514#8

Should Alice be allowed to give a clue that gets a card which is not the expected target in Cathy's hand because Bob knows the expected reaction makes no sense? If so, What should Cathy think about it?

Think about error recovery systems

When a card bombs, should ctd/kt cards be revoked if they are potentially newly critical?

Can deprioritization of a gotten target move focus from an unclued card to a clued card?

Solidify exactly when GDs do and do not promise position

Can a locked player give a stable lock clue to another player?

### Long term ideas

Revisit good touch
option: Good touch plays and gds are queued actions.

What is a bad stable clue?

Cold phase starts at score 10
Reactive clues target leftmost playable in Ava's hand

Conflicting Ideas:
"Don't skip over a player unless they have a safe action"
"The best stable clue should be legal for stalling players to give"

Proposal: Allow reactive targeting for delayed playables that depend on fully known cards in Alice's hand
-reverse reactive situations only?
delayed playables that depend on card not in avas hand could be allowed with lower priority?
allow play through fully known cards only?

Make the stack not be a queue, reactive clues resolve the round they are given?

If Alice is locked, it would be nice to have 1 expected stable clue to Cathy allowed which leaves Bob locked

Proposal: Stable clues that reveal a cross-hand dupe do not have any additional meaning

### Common errors

- Prioritizing fixing clued trash instead of targeting leftmost trash
- Not noticing a reverse reactive situation when lock plus KT applies
- Making a mistake with finesse targeting
- Forgetting to deprioritize gotten cards as initial signal targets
- Confusing hot/cold and skippable/reversible

Idea: If a reactive clue resolves unconventionally, it is conventionally considered cancelled. If Reuben goes back for it, Ava should follow.
https://hanab.live/shared-replay/1545676#29

## Discussion Notes
### Notes 6/24

Define our goals for each game
Type 1: Winstreak
Type 2: Practice
Type 3: Experiment

Lets set solid decisions for some of our conventional gaps and then hunt the streak

If Bob has a signaled discard which could be a bdr, Bob is not locked and cannot stall
(We agree this isn't the theoretical peak of hanabi conventions)

Global elim? For same hand dups that have been discarded?
NO ELIM

Discussing modes of errors:
False global elim
Wrong target in Cathy's hand (esp if a lower rank playable exists)

#### New agreement:
If Ruben has a reaction queued and receives another reactive signal, they are expected to respond to the new reactive signal and the old reactive clue is cancelled and becomes an optional action. If Ruben goes back for the first reaction, then Ava is expected to respond.

### Notes 6/25

Fill ins:
Filling in 1 away card and new card
play+discard if possible, but if cathy has no plays and no trash then bob targets the filled in 1-away
if connector plays, no discard signal for cathy
this helps us finesse into nearly locked cathy
dup extraction not allowed here, if no trash discard then bob targets the 1 away as a play

how does this work on a newly touched discard clue that looks like possible dup extraction on a crit
target leftmost non crit as discard

### Notes 6/26

If Bob has a queued play, Alice clues Cathy, and Bob discards the play, should Cathy believe it is a gd or a reaction?

Add a *Principles* section to resolve situations that come up without a convention agreement? One such principle follows:

Possible Principle: After Alice clues Cathy, Bob should only take conventionally expected actions unless he is okay with Cathy trying to interpret it as a reactive clue.

### Notes 7/2

Considered and rejected idea:
When Bob recieved a potential help-yourself play clue, He and Cathy could agree to deprioritize the next playable in that suit in Alice's hand for reactive clues that round. We decided not using this currently because help-yourself clues are not sufficiently defined.

For discussion:
https://hanab.live/shared-replay/1537460#7
How do we focus when some of the touched cards are already gotten? We allow initial signal to target gotten cards (at least as discards) so it's odd to ignore them when touched, but could allow more slot 1 reactions?

Can a player with a chop gives stable discards to Cathy if Bob has kt?

Should help yourself clues promise position of the playable if multiple cards touched

When are bad stable clues enough to cause a reaction? If Timo had y5 and Fafrd was locked, would y5 help yourself to timo be 'bad' enough?
https://hanab.live/shared-replay/1537725#14

### Notes 7/3

We like the idea of not focusing gotten cards because it's nice to put the inital target on cards which could be final target to get more slot 1 reactions.

New possible agreement: (Now adopted)
We ignore the 'gotten' status of cards when determining initial meanings for reactive clues and just decide based on empathy.

This gives up on some slot 1 reactions in favor of simplicity.

How strong is good touch?
Cross-hand dups allowed
If 2 blue cards, one has !4, do we believe b45 exactly? NO!

Proposal: Good touch applies to cards which are known playable or trash, which cannot depend on other cards also being assumed good.

Proposal: If the global empathy on a card is such that all of its identities are trash or playable, then it is considered queued for the purpose of reactive clues. If multiple such cards exist, they queue from right to left by default.

Problem: Imagine Bob has a queued y4 and Cathy has a clued yellow card. Is Cathy's yellow card queued from good touch? If so, then Alice can give stable clues to Bob.

Do we want to add 1s order meaning?

### Notes 7/4

Idea:
If Cathy is locked and Bob has any safe action, Alice can give reverse reactive clues to Bob.

To what extent does it matter if Alice is locked?

Designing for midgame and endgame are in conflict. In the endgame we want Alice to be able to decide she should stall, but in the midgame we care about whether Alice is locked to try to get more eficiency.

Vague idea: conventions could change at a set threshhold (score 10) to get the best of both worlds

Idea:
skippable = has a known safe action (play, trash, stable discard target)
reversible = has a play OR (has known discard and Cathy is locked)

Currently trash can only be skipped if Alice is locked

Good touch proposal: (adopted)
A card becomes queued from good touch the moment that it becomes immediately playable or trash from global empathy.
If there are more touched candidates than possible playables, none are queued without further help.

We have some concern that we can make too many trade-offs towards allowing less efficient clues in the mid game and the clue count will become an increasingly large problem.
When playing through easy midgames and optimizing endgames, it can be tempting to change the conventions which made the midgame strong without realizing the consequences.

Things that can be included in clue interpretation flow charts:
Is Alice locked?
Are we in the endgame?

#### Major convention shift:

Split the game into a "Hot Phase" and "Cold Phase" along the existing line of choplessness. Many decisions which involve trade-offs between efficiency and endgame considerations can be handled differently in these different phases.
This is not fully agreed to, and suggests many details to be filled in yet.

### Notes 7/6

The value of a discard signal largely depends on how many other actions are in that hand. If the discard is the only safe action, it is quite low value.

Should we be able to get chops on low clues? What if Alice has all good cards and we don't have enough clues to lock?

We want to make sure our cold phase/chop condition is easy to track and sync on

score >= 10, clue count >2, card draw can be chop

maybe clue count/number of unlocked players should be involved in deciding cold phase transition

idea: make rank clues also call for double plays in late game

deferring a crit play is rarely good, should hot clues apply?

Alice's status should matter at least somewhat to clue interpretation

### Notes 7/8

Previously:
Cold phase starts at score 10
Cards drawn during the cold phase are on chop by default (including the score 10 draw)

Hot Phase and Alice unlocked: Hot Clues
Cold Phase or Alice locked: Cold Clues
Cold Clues:
- skippable = has a known safe action (play, trash, stable discard target)
- reversible = has a play OR (has known discard and Cathy is locked)
Hot Clues:
- skippable = has queued play while other player doesn't
- reversible = has loaded action

Does fill-in focus apply to queued plays?
It doesn't help get slot 1 reactions
It feels unintuitive, maybe not even better for sync
Proposal(adopted on trial): Initial reactive target deprioritizes queued plays -----
Should this include kt? What about potential bdrs?

When is a delayed playable card targetable?
Proposal(adopted): Delayed playable cards can be targeted iff the intervening card(s) are all in Ava's hand -----
-rewrite fill-in focus to only apply to cards which are valid reactive targets

Idea queued to discard (qtd) terminology

Hot and Cold clues
Should Alice's status be considered?
Proposal: Having a play/crit play makes Alice's clues hot
Proposal: Bob is reversible whenever he has a play. He is also reversible when he has known trash and Alice or Cathy is locked

Alice is cold = Alice is locked or it's the cold phase and Alice does not know about a play
Proposal: Bob is reversible whenever he has a play. He is also reversible when he has known trash and Alice is cold or Cathy is locked
Proposal: A player is skippable whenever they have a queued play and the other doesn't. A player is also skippable if they have known trash and Alice is cold

Stacked reactive clues
Proposal(adopted on trial): When Alice is hot, the first player out of Bob and Cathy to run out of plays is Reuben and the signalled actions are added to the end of Reuben and Avas' queues.
- Targeting in Ava's hand is based on play stacks at the time that Reuben runs outs of plays and can interpret the reactive clue.
-Concern: How do we stay in sync about queue length? How does error recovery work here? Can a player stop to fix? What if Cathy has 2 known plays and doesn't know the expected order for them?
-Follow up: precise definition of what counts as queued
-Follow up: reconsider if we like good touch queue timing for this purpose?
- Clues to Reuben are undefined? Stable?

Good touch?
bad stable clues?
do gds promise position

Proposal: Bombing a queued card with only one playable identity unqueues any queued playable with only that identity in all other hands.
Relevant turn: https://hanab.live/shared-replay/1542481#23. Discarding here and then playing g4 causes
Update: See https://hackmd.io/H5OrlGYTT1y_OoBchjuK9w?both#Good-Touch-Critical-Bomb for the actual wording we adopted.

### Notes 7/12

Should gds allow players to act on private knowledge?
Should players be allowed to gd instead of playing as a reaction if the card is exactly known (like from a finesse)

If Bob plays early from a reverse reactive clue, treating it as stable, do we want it to cancel Cathy's reaction? yes

We could consider making critical cards kigher priority targets than non-crits

Bad stable clues get in the way of fix play+discards, particularly if the stable meaning would be dup extraction

### Notes 7/13

Variant idea: Dead Suit
The dead suit is like a normal suit (maybe dark?) but cannot be played or discarded
https://hanab.live/shared-replay/1546476

### Notes 7/15

Proposal: Cancelling a reverse reactive clue by treating it as stable should require taking stable meaning besides just globally known empathy.
https://hanab.live/shared-replay/1547792#8
Here if Timo clues 2, playing an empathy playable 2 shouldn't cancel the reactive signal

Proposal: A playable rank clue play signals the focused (leftmost newly touched) card, and carries no additional meaning besides good touch for the other touched cards.

Should too many 1s prevent them from being queued?
Note that 1s are queued when playable, and if extra 1s are clued later that does not unqueue the older ones

It would be nice to have a good touch play order clearly defined to know when a fix is needed

### Notes 7/29

It would be nice if this was legal
https://hanab.live/shared-replay/1559284#26
Idea: If a stable clue counterfeits a playable which discards, it isn't interpreted as a reaction?

### Notes 7/30

When considering whether players are skippable or reversible, does 'locked' mean exactly locked or just has no known action? When we have chops, Cathy won't know about her action until her turn when she has nothing else to do, and our current terminology may not account for the difference properly.

Proposal: Stable clues that reveal a cross-hand dupe do not have any additional meaning

### Notes 8/6

We have had a few times that we tried to treat signalled discards as deprioritized for new clue focus/targeting, but the current agreement says that is for queued plays only. Should it be expanded to include discards?

### Notes 8/7

We are not thrilled with hot and cold clues as they currently exist. They haven't become intuitive with some experience and we are frequently needing to check the doc and interpret the confusing exact wording to try to sync on answers. It would be nice to develop a cleaner configuration for them.

Timo would like to revamp the cold phase

Alice cold, cold phase, bob loaded, cathy has chop
Alice wants to stable play clue bob, would like the cancelling reverse reactives idea

Sodium mentions that we could also benefit from making rank clues double plays. Revamping the cold phase and reconfiguring reactions can go together.

We also consider formulations where we make it Bob's job to not confuse cathy when a potentially confusing scenario arises

### Notes 8/13

We seem to be caught between worlds in our error recovery system. We like to sync to the mistake by demonstrating where our conventional status is, but we also want to be able to fix a known mistake and resync to how things ought to have gone.

### Notes 8/15

Does GD's promising position depend on a card being immediately playable, or are delayed playabled promised as well? Which way does playable through a player's own hand go?

When exactly does a clue become reverse reactive?
If Bob is both skippable and reversible then clues to Bob are stable? Does it depend on Cathy being skippable?

### Notes 8/18

Can a locked player give a lock signal to another player?
https://hanab.live/shared-replay/1575799#30
Is this a discard signal because sodium is already locked, or is it a stable lock clue as normal?

If reactive rank and color were swapped, cold alice giving stable discards to cathy would be nice

### Notes 8/26

Some confusion about expected actions allowing reverse reactions vs unexpected actions cancelling them. Particularly, if playing an elim card could allow a stable clue to become a reverse reactive. Currently it could not, but Timo did not realize this implication to our current agreement.