← all games
Hanabi

Hanabi

#597BGG ↗

2010 · 2-5 players · 25min · weight 1.69 · 52,397 ratings

port: BGAdifficulty: Hard
BGGv4 widev4v4 deep5Sources7Rules cardCandidateDeep dive
Bayes
6.92
Users rated
52,397
Owned
90,390
Wishing
4,728

At a glance — v4 wide

Controlled-vocabulary primitives + 8-axis MDA aesthetic vector. Vocab v2.

Core loop (micro)

Hold cards facing away; give a color or number clue to a teammate so they know which card to play next without seeing their hand.

MDA aesthetic vector (0–3)
Sensation
0
Fantasy
0
Narrative
1
Challenge
3
Fellowship
3
Discovery
1
Expression
2
Submission
1
Mechanic-interaction primitives (4)
  • [3]communication_constraint— “limited info-sharing during play — clue tokens limit how many hints can be given
  • [3]info_asymmetry_stable— “hold cards facing teammates; each player permanently lacks info about their own hand
  • [2]forced_table_talk— “game requires verbal coordination; players must coordinate clues without full communication
  • [2]attrition_clock— “fail token economy — three misplays end the game; fuse tokens deplete on wrong plays
discovery_score: 0.011

Archetype fits — v4 deep

How well this game shape maps to mobile archetype templates. Composite is a weighted sum of the 10 fit dimensions.

ArchetypeCompositeLTFSessionComboArcShare5inOnboard
Coop
Hanabi is the canonical coop info-asymmetry exemplar. Rules excerpt confirms the strict communication constraint ('one, and only one, of the following actions') + v4-wide info_asymmetry_stable + semantic_overloading + communication_constraint all at strength 3 — bluff_info_asymmetry pegs at 9 (the dimension's highest signal of coop-fit per the rubric). Visual legibility strong at 8 (Hanabi-shape: 4 cards visible to teammate + piles).
7.208.0978587
Balatro
Hanabi has hidden-info structure (own hand unseen) but no scaling combo / engine growth — fireworks rows are linear 1→5 builds. Wrong shape for Balatro's run-based escalation.
5.205.0646576
Wordle
Single-puzzle resolve shape but session is too long (~25min vs Wordle's 60-90s) and is fundamentally multiplayer-coop, not single-solver.
5.007.0536575
Cozy
Coop-or-Solo tier (Cooperative Game) — no cap. Tonal art direction (Ralenti watercolor fireworks) is cozy-adjacent, but cognitive load of memory + clue-tracking is decidedly not low-key. forced_table_talk primitive at strength 2 is anti-cozy in feel even within Coop tier. Score against tone: 5-7 range.
4.907.0535476
Snap
Information asymmetry exists but in cooperative direction — no PvP factional deck, no lane structure. Wrong loop.
4.605.0435475

Rules card

Synthesized from sources below. Readiness: ready. Confidence: 1.00.

Readiness

ready (confidence=1.00, rules=1.00, fun=1.00). BGG rank: 597; year: 2010; weight: 1.69; playtime: 25 min

SourceQualityRoleNote
pdf1.00rules authorityBGG file 86970 (English from French translation, 1 page)
llm_memory0.90draft synthesissonnet-self-rated-9
bga_tutorial0.85rules authorityBGA implementation rules summary
bgg_comments0.75player voicepositive/player-voice sample
youtube_transcript0.70teach-flowhow-to-play transcript
wikipedia0.55context/receptionboard-game-suffix
github_code0.15implementation signalGitHub match is adjacent/tooling, not rules code

Core Loop

A fully cooperative card game where players hold their cards facing outward — you cannot see your own cards, only others'. The shared goal is to play fireworks cards in sequence (1-2-3-4-5) for each of 5 colors. On your turn you take one of three actions: give a clue, play a card, or discard a card.

Giving a clue: point to one or more cards in another player's hand and tell them either "these cards are all [color]" or "these cards are all [number]." You must identify ALL cards matching that attribute — you cannot give a partial clue. Spend one of the 8 (or sometimes less) clue tokens from the shared supply.

Playing a card: choose a card from your hand and play it to the communal fireworks display. If it's the next number in its color's sequence, it's placed successfully. If not, it's a dud — discard it and lose one of 3 fuse tokens.

Discarding a card: remove one card from your hand and gain back one clue token. Draw a replacement card from the deck after any action.

The team relies on shared knowledge — players use clue logic to infer what their cards are, and must act on incomplete information.

Turn Structure and State

  • Publisher/manual source is present and treated as the top rules authority.
  • BGA tutorial is present; useful for exact turn flow and implementation gotchas.
  • How-to-play transcript is present; useful for teach order and confusing steps.
  • BGG description anchor: Hanabi—named for the Japanese word for "fireworks"—is a cooperative game in which players try to create the perfect fireworks show by placing the cards on the table in the right order. (In Japanese, hanabi is written as 花火; these are the ideograms flower and fire, respectively.) The card deck consists of five different colors of cards, numbered 1–5 in each color. For each color, the players [...]

Win Condition and Arc

Win by completing all 5 fireworks sequences (1-2-3-4-5 in each of 5 colors) for a perfect score of 25. Score equals how many cards were successfully played when the game ends (either deck exhausted + one final round, or all fuses blown = immediate loss with score 0 or current points). Lose immediately if 3 duds are played (3 fuses blown).

The arc builds around trust in clue interpretation. Early game is establishing a shared "protocol" for how clues mean what; late game is high-stakes plays with few clue tokens and known fuse pressure.

Decision Primitives

BGG mechanisms: Communication Limits, Cooperative Game, Hand Management, Memory, Ordering, Set Collection

Memory-derived primitives:

  • Hidden information (can't see own cards)
  • Cooperative communication with constrained clues
  • Shared resource management (clue tokens, fuse tokens)
  • Inference and deduction
  • Hand management under uncertainty

v4 controlled primitives: communication_constraint, info_asymmetry_stable, semantic_overloading, forced_table_talk, attrition_clock

Top iOS archetype fits: coop 7.2, balatro 5.2, wordle 5.0.

Why It Is Fun

The core tension is exquisite: the game is hard enough that perfect scores feel like genuine achievements, but the constraint of only saying color or number (never "play this" or "don't play that") forces a creative compression of information. The joy comes from the team successfully inferring complex states from minimal signals.

Player-voice evidence:

  • I really enjoy the silent mechanic of this game. It's a set-building co-op of deduction and memory. I'm a strict player, and if I could I would make everyone play in total silence, and I would love every tension filled second. BUT you...
  • great coop, where everybody works together, without one person controlling the gameplan. Easy to understand, plays quick
  • Quick, sleek, and fun filler. At its best, a rare co-op that can't be managed by an alpha gamer in the group, provided players don't start mistaking metagaming for logic.
  • Really fun, and challenging deduction game. Each player does have to pay attention though, and be pretty switched on. If you want a game that makes you think, and work with everyone to win, this is for you.
  • Really love the group dynamic that it creates. Fun strategy and tactics aswell. Also always striving to do better next game.

Friction and Failure Modes

  • Treat Sonnet-memory edge rules as draft until confirmed by manual, BGA, or transcript.
  • GitHub source is adjacent/tooling or false-positive; ignore for rules semantics.

Translation and Design Hooks

  • Use this card to ask: which primitive carries the fun if theme/licensing is removed?
  • For iOS, look for short-session compression, clear state visualization, and a digital-only twist.
  • For new tabletop design, look for the tension source and decide whether to preserve or invert it.

Edge Rules and Gotchas

  • Clues must be truthful and complete — you cannot point to only some cards of the matching attribute; you must indicate all matching cards even if revealing too much
  • When the deck runs out, each player gets exactly one more turn (including the player who drew the last card), then the game ends and score is tallied
  • If a 5 is successfully played in any color, you immediately gain 1 clue token back as a bonus
  • Multicolor variant adds a sixth color (rainbow) that matches every color clue, dramatically increasing difficulty
  • The "good hand" convention: experienced players often use the position of cards in hand as implicit meta-information (leftmost = oldest, newest cards may be unclued but playable soon)
  • Discarding a card you still need permanently reduces max possible score; careful tracking of discards is essential

Sources Used

[
  {
    "kind": "pdf",
    "path": "data/rulebooks/98778.pdf",
    "quality": 1.0,
    "note": "BGG file 86970 (English from French translation, 1 page)"
  },
  {
    "kind": "llm_memory",
    "path": "data/llm_memory_sonnet/98778.md",
    "quality": 0.9,
    "note": "sonnet-self-rated-9"
  },
  {
    "kind": "bga_tutorial",
    "path": "data/bga_tutorials/98778.md",
    "quality": 0.85,
    "note": "BGA implementation rules summary"
  },
  {
    "kind": "bgg_comments",
    "path": "data/bgg_comments/98778.txt",
    "quality": 0.75,
    "note": "positive/player-voice sample"
  },
  {
    "kind": "youtube_transcript",
    "path": "data/youtube_transcripts/98778.txt",
    "quality": 0.7,
    "note": "how-to-play transcript"
  },
  {
    "kind": "wikipedia",
    "path": "data/wikipedia/98778.md",
    "quality": 0.55,
    "note": "board-game-suffix"
  },
  {
    "kind": "github_code",
    "path": "data/code_implementations/98778.md",
    "quality": 0.15,
    "note": "GitHub match is adjacent/tooling, not rules code"
  }
]

Sources (7)

Inputs to rules-card synthesis. Click any pill with ↗ to open the original source.

Rulebook PDF1.00LLM memory0.90BGA tutorial0.85BGG comments0.75YouTube transcript0.70Wikipedia0.55GitHub code0.15

BGG tags