# Reactor 3

## Central Features

### Stable Clues

Color to a new card is direct play
Rank to a new card is discard to the left
Fill-in focus applies to reclues
A discard signal to chop is actually a lock signal
A discard signal to Bob's slot 1 on turn 1 is actually a lock signal

#### Help Yourself Clues

Stable color play clues to Reuben which match a suit that Ava could play on ask Reuben to get Ava's card(s) in that suit and play his own after. Reuben is locked until his card becomes playable or he receives another signal.

If multiple cards are touched with a HY clue, the focused card is not immediately promised to be the next playable. Reuben should not attempt to discard anything right away as if the identities are locked in. The position of the playable is not promised until the card is immediately playable and it is Reuben's turn.

#### Fill-in Focus

We allow stable clues to focus on a filled-in playable or trash and have no other meaning.

### Reactive Clues

Default reaction is in limbo, probably will still be slot 1
Reuben counts slots in the opposite direction from Ava

Initial and expected targets are in limbo

### Hot Phase & Cold Phase

In the early game, we don't want to spend clues to lock players, as their starting hands are often full of good cards. We say that up until the score hits 10 (number subject to change soon), players have no chop slot and are locked by default. All discards in the hot phase must be signalled. The card drawn by the 10th play is the first chop, and chop is the leftmost card after that. Note that each player must draw their first chop in the cold phase; no card that was present in the hot phase is ever on chop.

Hot and cold can also affect clue interpretations, but that is currently being discussed.

### Good Touch

While we don't ignore good touch, we do treat it somewhat softly, as the immediate value of a bad touch clue can be high and it is often possible to give fixes efficiently. It is then important to be synced on the global status of any potentially good touch playable cards in order to interpret new reactive clues.

## Additional Features

### Error Recovery Procedure

### Cross-hand Duplicate Extraction

### Known Duplicate Targeting

### Negative Fill-Ins

### Rank Stalls

## Recent Changes

### Expected Deductions
Cards which would be good touch playable that are proven to be not playable by reactive targeting rules + Reuben's reaction are expected to be marked as trash.

This is evaluated when Reuben reacts, Ava is not expected to figure it out later or after her own signalled action. Ava is also not expected to reason about what reactions may be impossible. The team should mark trash as though any reaction was possible for Reuben.

## Notable Exclusions

### Bluffs

Bluffs can work in theory, but they invite desync. Alice attempts a bluff at her own risk, her teammates are not responsible for accounting for any possible bluff interpretations.

### Elimination

Elimination may be formally included in a limited form at some point. The problem is that it needs to be fully globally synced in order for reactive clues to work properly, which is very error prone. Additionally, any mistake can cause false elimination notes, leading to desync escalation. Instead, we say that elim is fully optional for each player and never affects clue interpretations.

### Priority

No

### Trash Pushes

A stable clue which newly identifies trash has already communicated a safe action, we don't expect it to do more. This is most relevant in difficult decks where there are now many safe actions available and we want more options for how to get the few good actions.

### Shout Discards

Players are welcome to take a signalled discard before a play without it meaning anything special.

### Trash Order or Play Order

We usually discard known trash from right to left, but have assigned no meaning to discarding out of order. We allow players to use their own judgement and any context they may have for choosing what order to play their cards or take their signalled discards.

## Ideas

### Design Goals & Principles

We would like
1 giving reactive clues and being efficient often
2. stalling when it's good for endgame
3. a stable clue always available, even when bob is locked

### Current Topics

Hot & cold framework

Color and number swap

bad stable clues

error recovery
revoking ptds after a bomb
weakening elim

### Future Topics

deprioritizing reactive targets
fully known cards
multiple copies of a playable
known trash
can deprioritization of a gotten target move focus from an unclued card to a clued card?

having multiple locked players
closely related, having some stable clue always available

#### Ideas to Revisit:
good touch and gds
current good touch agreements include many special rules, would like to be unified
cold phase threshold
we all seem to want >10, maybe 12?
Reactive clues targeting leftmost
This conflicts with being able to plan ahead beyond the next draw
double discard vs finesse
stacked reactive clues

#### Fafrd's Questions
does focus inversion exist? -No
how does color direct dupe extraction expected targeting work? (this question may well become obsolete soon -slot 1 reaction preferred

#### List of things we aren't currently changing
reactive clue counting
stable direct color clues
HY clues to bob

#### The Stable Configuration

#### The Reactive Configuration

Design goals:
- Get better clue predictability: deprioritize slot 1 actions and slot 1 focuses.
- Better distinguish double plays from good stables.
- Continue to get fill-in finesses (in otherwise locked hands)

## Current Experiment Parameters (2/6/2026)
- Stable color clue target precedence is: 15432
- Expected reactive target priority 23451, focus priority 34521, initial target is left-referential (ignoring slot 1)
- no-info reclues are p+p, fill-in reclues are p+d
- initial/expected reactive meanings ignore slot 1 unless its the only one touched