---
tags: Hanabi
---

# Turbo - a Hanabi convention system
The official Turbo reference has moved: see the [Hanabi Central Wiki](https://hanabi.wiki/en/conventions/turbo) for the latest conventions.

## Introduction
There are two major equivalent ways to think about this convention system. The first is that it's what happens if we start with Referential Sieve and then strengthen the Sieve Principle, demanding that no drawn cards can ever bypass the sieve. The second is that this is just a discard oldest system that has a tremendous number of non-permanent chop moves (most cards in the starting hand and then some).

### Acknowledgements
**sjdrodge** - Sieve Principle via Sieve Conventions and Strong Sieve Principle via this system
**pianoblook & piper** - Referential Principle via the Good Trash System
**timotree** - putting Sieve Principle and Referential Principle together to make Referential Sieve
**Libster** - helping with Strong Sieve Principle
**Floriman** - the idea that discard oldest + starting hand chop moves might be a strong alternative to sieve systems

## Basic Principles

- **Loaded Principle** - A player is considered to be "loaded" if they have a known safe action (i.e. known trash or a known playable card)
- **Strong Sieve Principle** - every drawn card enters a queue. The card at the front of that queue is said to be "on chop." If you are given a play instruction while "unloaded," then the card at the front of the queue "passes through the sieve" and the next card in queue is now "on chop." If you are given a discard instruction, multiple cards may "pass though the sieve" at the same time.
- **Referential Principle** - Instead of directly cluing the card which is receiving a play/discard instruction, we prefer to clue a card that will remain in the hand. Therefore, most clues in this system are "referential" in nature.
- **Safe Action Principle** - Any fill-in clue given to an unloaded player which demonstrates a re-clued card is an immediately safe play/discard is considered valuable enough on its own and therefore doesn't carry its usual referential meaning. Such clues cause the card currently "on chop" to "pass through the sieve."
- **Discard Precedence Principle** - If you receive a discard instruction, all cards with higher discard precedence than the one you are instructed to discard "pass through the sieve."

## Basic Conventions

Conventions in this section are assumed to be on; they have a proven track record.

### Discard Precedence
The card on chop has the highest discard precedence. Precedence then descends through the unclued cards to the left and wraps around through the unclued "chop moved" cards.

### Referential Discard Clues (w/ rank)
Rank clues give a discard instruction to the card to their left. If you give a discard instruction to the card with highest discard precedence (chop), it locks that player instead. [TODO: clarify when multiple cards are involved]

### Referential Play Clues (w/ color)
Color clues give a play instruction to the card to their left. [TODO: clarify when multiple cards are involved]

### Direct Play Clues (w/ rank)
If all unplayed cards of a given rank are playable, a clue of that rank is a direct play clue instead of discard clue.

### Fill-in Clues
Any clue which fills in a previously clued card and turns it into a safe action (trash or playable) is not a referential clue.

## Experimental Conventions

Conventions in this section are being tested. There's no guarantee that they are mutually consistent.

### Starting Hand Discard Precedence
Chop begins on slot 1 (all other starting hand cards are "chop moved").

### Mirror Play Clues

Color clues focus the rightmost newly-touched card, providing a play instruction on the the card the focus reflects to through the previously-touched cards.

### Unbalanced states

An *unbalanced state* is one where one player has significantly more safe actions than the other. The main way this can occur is when a player without any safe actions is able to give highly efficint clues to their teammate (no-infos, trash pushes). While ascending the hump, without special conventions to handle them, unbalanced states can feel bad to enter. It can be impossible to save cards entering the more loaded player's hand, and it can be more expensive to lock the less loaded player's hand. This higher lock expense incentivises the less-loaded player to give underinformative clues. This section exists to mitigate that.

#### Early Save Clues

Suppose Bob is loaded with (future) chop currently on or rightward of slot 2. Then Alice's rank clues touching cards left of (and including) chop are early save clues, referring leftward and biased on the rightmost in the sieve queue. Not saving all cards in the discard queue can provide Bob with a safe discard (*early permission to discard*). With permission to discard, Bob can lock Alice more easily.

#### Scream chop move / (lock!)

If Alice has a play and a discard, she can chop move one card in Bob's hand by taking the discard. When playing with discards being expected from the right, discarding a leftward known discardable could chop-move multiple cards (is this h-group?).

### Early Game Adjustments

#### Starting Hand Chop Move Timing

A player's starting hand is chop-moved after they take/receive their first action. Then, ptd will be given to the rightmost card, starting hand stalls touch the leftmost card in hand, and starting hand discard signals refer left with a right-bias. OCMs will index in from the right.

This, in my (sodium's) opinion, is a simpler agreement than cm'ing all but the leftmost; in the alternative, OCMs (discards wrapparound to the rightmost?) and starting hand stalls (apparent discard clue touching slot 2?) in particular are less intuitive in such a situation.

#### Clue type swap

For the first 2 (4? 6?) turns of the game, swap clue meanings: rank clues are referential plays, color clues are referential discards / starting hand stalls.

#### 4s on chop (2p)

If a 4 is touched and is (global/empathy) known to be noncritical 2+ away from playable, discarding it (the leftmost of these) is the chop action. Taking this discard for free can lead to

- Better clue count maintainance, leading to more easily saving other good cards and providing ease in ascending the hump.
- Some double discard avoidance

Because this convention is based in a claim about the use of these good cards, this chop is harder to cm: a card would have to play/discard that makes at least one possible (global) identity of the 4 2-away or critical.

### Low clue states

At sufficiently low clue states, revoking permission to discard better have a good reason: it should either be a double discard situation or it should promise a safe way out.

#### 0-clue safety promise

When Alice gives a play clue to enter a 0 clue state, she usually promises a safe discard in Bob's hand. The card is the rightmost untouched not-cm'd card in Bob's hand. When this card exists, the situation is generally *safe*. When the card does not exist, the situation is generally *unsafe*.

#### Safe 0-clue States

In a safe 0-clue state, when Alice's (future) chop is not known to be a safe discard, Bob is expected to scream discard for it. Not discarding immediately promises Alice a safe discard if she does not already have one.

#### Unsafe 0-clue States

When Bob is not promised a safe discard and Alice only has one discard, the situation is unsafe. Players should be scared. There should be an agreement on what card will discard if players are unable to play into each other, and that discard can occur early if a player sees that the one in their teammate's hand is bad. Generally, unsafe situations will resolve by either
- The team hoping that at least one of two cards is safe to discard (players transform into gnomes in hat problems).
- The team unexpectedly plays a 5.

#### Exceptions to the safety characterization:

- 5s to play
- Bob doubly-loaded with Alice holding extra trash
- Discard modulation opportunities