hanabi

@hanab

Public team

Joined on Dec 27, 2022

  • Current Winstreak: 12 https://hanab.live/shared-replay/1529187Desync on turn 20: Fafrd thought g3 elim would be global and make sjdrodge loaded. https://hanab.live/shared-replay/1529216 https://hanab.live/shared-replay/1529606 https://hanab.live/shared-replay/1530483
     Like  Bookmark
  • Don't worry, this is going to be a real document soon. no turn timers chat in zen mode no voice call no dms no external communication of any kind no external tools (specify some deets) Active Ethics
     Like  Bookmark
  • Reactor AT DAWN is an experimental variation of Reactor 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, 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. 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. 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.
     Like  Bookmark
  • Note: The focus priority here may be out of date for how we have been playing. See https://hackmd.io/H5OrlGYTT1y_OoBchjuK9w#How-are-reactive-clues-typically-focused initial + 1 = reaction + target In order of focus priority. Type of Clue Prior Clue on Focus Signaled Actions Initial Target
     Like  Bookmark
  • On 2025/4/29 timotree and Fafrd begin winstreak attempts. We will use this document to record convention ideas from games, rather than inventing conventions midgame. Current streak: 3 0 Conventions 2p ref sieve. 7 clue permission to playdiscarding to 8 clues when partner has no queued plays gives a play signal if Alice's only known actions are discards and Bob is unloaded, then she is in a Permission to Play situation and clues to Bob which normally mean play slot 1 instead mean discard slot 1 playing last action at 8 clues gives permission to bomb; note says that slot 1 is noncritical unless playable
     Like  Bookmark
  • Fafmark We have a few critical, base convention special agreements. First, our common aggreements are: Rank Trash Promise Negative Fill-Ins Next, the changes to base convention interpretation: Elim Rider Deduction / Safe Action Principle (Shared?) If a possible referential discard clue reveals a play for any reason, then it can't also give a discard signal, except through good touch or 0cs agreement. A possible referential discard clue is an unloaded clue with rank that touches at least 1 new card.
     Like  Bookmark
  • PurpleJodium No Variant 1s Order: H-Group 1s should play in the following order: Chop Fresh Right-to-Left Playing out of order may trigger something.
     Like  Bookmark
  • CA Exports This is the 2p convention agreement page for hallmark and sodiumdebt. We have played 2p ref sieve together since 2022. All standard agreements. We use the following common agreements: PTD Contract Quasiloaded PTD Pre-Prompt Bluffs
     Like  Bookmark
  • Nafrd All standard agreements. We also use the following common agreements: Pre-Prompt Bluffs Revoking PTD If we revoke PTD with a color referential play clue, then at least one of the following holds: The revoked PTD is newly-critical this round. The revoked PTD is newly-playable (possibly delayed) this round.
     Like  Bookmark
  • Macmark No Starting Hand 1s Agreement We do not play with a starting hand 1s agreement, so fill-in clues that may be prompt or finesse soon after turn 1 should be a prompt if possible.
     Like  Bookmark
  • Introduction Reactor is a system (currently!) only for three player consisting of two types of clues: Referential Sieve clues to the next player needing a safe action (the reacter) Reactive clues to the other player (the reciever) The Referential Sieve clues tell the reacter about their hand, while the reactive clues provide a safe action action to both players. Reactive color clues provide a play and a discard. Reactive rank clues provide two plays. The Reacter For the purpose of a reactive clue, the first player to take an action indicated by the clue is called the reacter.
     Like 2 Bookmark
  • Here, we document the agreements between different teams within 2P Referential Sieve. Table of Contents Players/teams are listed alphabetically. Players Fafrd Hallmark PurpleJoe Sodium
     Like  Bookmark
  • Table of Contents Book
     Like  Bookmark
  • image
     Like  Bookmark
  • Starting hand 1s Order: Left-to-Right 1s should play left to right. Playing out of order may trigger something. This is a standard agreement. 1s Order Chop Moves Playing 1s out of order chop moves slot 1, shifting the expected discard inward by as many non-chop-moved cards as were skipped in the 1s order. This is a standard agreement. 1s Order Loaded Discards When Bob is loaded, Alice playing 1s out of order provides him a discard, counting inward as many cards as were skipped, starting from slot 1. Currently used by Fafrd / Sodium, PurpleJoe / Sodium
     Like  Bookmark
  • This is Steve. Steve is being crafted to balance several priorities: Tempo: we desire play signals to be given early to avoid unnecessary locks without losing delayed playables. Efficiency: we use referential clues to optimize information gain. Discard quality: we want to be able to save any card as needed. Discard planning: we want to see chops to be able to best prepare for them. Play signal planning: we want to be able to predict future play signals. Context: we want to reason effectively about our hands, consistently syncing with regards to focus inversion. Some of these priorities synergize. For instance, tempo aids context. They can also clash: clue space dedicated towards optimizing discard quality could otherwise be used to improve play signal diversity.
     Like  Bookmark
  • This document compares the most common 2p systems, Referential Sieve and H-group. I analyze their relative strengths with the explicit purpose of proposing design goals for a hypothetical system to incorparate power from both. In my ideal world, I'd like these analyses to be community efforts. But due to Hanabi's complexity, it is impossible to prove objectively what is a strong convention: in fact, as far as I'm aware, "strong" doesn't even have an objective meaning when applied to an individual convention. In lieu of proof, we can only theorize, with our evidence being hueristics backed up by games played. In addition to limiting knowledge, lack of proof means individuals often disagree, limiting even shared community beliefs. That said, it's still worthwhile to have our understanding documented. Although this work is entirely my point of view, I would enjoy converting it to (or starting a new) a community document given sufficient interest. As part of that, I would expect some (many) of my thoughts to be overidden due to others' disagreement. With that said, let's get into the comparison. Advantages: Referential Sieve These are traits with clear benefits included in Referential Sieve but not in H-group, ordered roughly by how clear the benefit is. Tempo Clue Focus
     Like 1 Bookmark
  • Referential Play CLues
     Like  Bookmark
  • Introduction The original reactor document is here. In this document, I'll present a full system: if you're new to reactor, you shouldn't necessarily need to reference the original document. With that said, what's new? Conventional Improvements Addtional safety in 1-for-1s.We reintroduce the direct color play signals of sieve. We allow delayed play signals not promising the position of any connectors. Directly attributable to timotree. Good touch information given by reactive clues.
     Like  Bookmark
  • The biggest difference between designing conventions for 2-player and 3-player is the possibility of reactions. A clue to Cathy can have a meaning for Bob which is based on the contents of Cathy's hand and a meaning for Cathy which is based on the way Bob reacted. When designing a convention system with reactions, the key questions are: What clues from Alice get reactions? What actions from Bob are considered reactions? What does each possible reaction from Bob tell Cathy? Viewed in this frame, Hat Guessing takes the extreme position of saying that every clue gets a reaction and that almost actions from Bob are considered reactions, with the only exception being playing already known playable cards. In terms of what the reactions tell Cathy, it's usually that a particular slot is playable or trash. H-Group has a few categories of clues which get reactions
     Like  Bookmark