log

First public release of game titles on the Parlourite board game engine

Parlourite's game titles are now public. Notes on OpenSpiel cross-validation, determinism via Event Sourcing and CQRS, the game-definition DSL, and upcoming work on avatars, rulebook generation, and AI research use.

After about a year of development, Parlourite’s game titles are now public.

Until now the game list only ran internally. Releasing while the engine’s expressiveness, legal move generation, terminal detection, or rule description were still shaky did not feel worth doing. Visuals can be fixed later. Logic errors become harder to fix as layers accumulate above them.

Game definitions have been cross-validated against OpenSpiel for a subset of classical games. All published definitions are independent implementations based on this project’s own design. Respect to the prior projects in the General Game Playing field.

Initial lineup

The first lineup spans classical abstracts, traditional card games, dice games, and original titles. All running on the same engine and the same DSL.

Classical and traditional

GameCategoryImplementation notes
BackgammonDice + raceCombinatorial move expansion from dice, bear-off condition, pip count
CheckersClassical abstractMulti-jump enumeration, forced capture
International DraughtsVariant10×10 board, flying kings, maximum-capture rule
Micro Shogi (3×4 drop-shogi)VariantSmall board with drops, termination on short games
Mancala (Kalah)Classical abstractSowing with extra turns, capture condition
Nine Men’s MorrisClassical abstractPhase transitions (placement / movement / flying), mill detection
HexClassical abstractBoard connectivity, union-find expressed in the DSL
GomokuClassical abstractLine detection, optional forbidden-move rules
ReversiClassical abstractFlanking detection, flips in all directions
Dots and BoxesClassical abstractEdge ownership, extra turn on box completion
NotaktoVariantMultiple concurrent boards, misère condition
Tic Tac ToeClassical abstractMinimal DSL validation case
Cube LineClassical abstractAlignment detection in 3D
HeartsTraditional cardTrick-taking, heart-breaking, shooting the moon
CribbageTraditional cardComprehensive scoring patterns, crib handling, pegging
Crazy EightsTraditional cardWild-card suit-nomination state
Go FishTraditional cardHidden information, inter-player queries
Liar’s DiceDice + bluffHidden information, bidding history, challenge resolution
YachtDiceCategory scoring, category-selection optimization

Original titles

The original titles were designed to probe how far the engine can absorb specific mechanic categories. Overlapping categories are intentional, meant to stress a given family from multiple angles.

TitleCategory
Street Stall EmpireArea Majority
Voltage Surge: Grid ControlPush Your Luck
Cloud Fleet AcademyCombat Movement
Ten IncensesCooperative
Abandoned Theme ParkPush Your Luck
Golems’ FactoryAction Programming
Cooling RacksPush Your Luck
Jungle Instinct: Hidden OasisCombat Movement

The fact that this range runs on the same substrate is itself the core claim of the project.

Technical foundation

Determinism

Determinism was a hard requirement from the start. The same initial state and the same input sequence always produce the same board and the same event stream. Randomness is included in that guarantee and is reproducible.

The architecture that made this feasible was Event Sourcing and CQRS. Every game action is recorded as an immutable event, and the board state is always derived by folding over the event stream. The choice was appropriate. Replay, debugging, synchronization, and verification all become straightforward on top of this foundation.

Game-definition DSL

Rather than writing a separate engine per game, Parlourite uses a DSL to describe games declaratively, and the engine interprets it.

Fitting classical abstracts, card games, and dice games into the same DSL was within initial expectations. The hard part was reaching expressiveness sufficient for Eurogames: hidden information, simultaneous moves, economy mechanics, card-driven play, engine building. Generalizing these at the DSL level felt reckless at the outset, but the goal is now in sight. A long road remains.

OpenSpiel comparison

Claiming correctness solely on self-written tests is weak. External ground truth matters. OpenSpiel served that role, with legal move sets and terminal conditions reconciled game by game.

Reaching the level required to pass this comparison took a long time, and the result is grounded confidence in the engine’s expressiveness.

Known gaps

Known issues remain after release.

  • UI: event logs and legal moves still appear in mechanical form in places. Natural-language rendering will be addressed from both the definition side and the generation side.
  • Online play: synchronization, disconnect, and reconnect behaviors only surface under real usage.
  • LLM player quality: quantitative evaluation of whether the model picks sensible moves from the legal set is still shallow.
  • Persistence: known bugs remain in account, replay, and settings storage.

With the deepest layer in place, repairs above can proceed on this foundation.

Planned work

Alongside stabilization, the following are planned.

Avatars

Parlourite is designed so that anyone can try it in two clicks without an account. The priority is “try immediately when curious” and “easy to recommend to others,” and the Passkey-based account creation follows the same stance of not asking for names or email addresses. In this setting, a lightweight auto-generated avatar is introduced so players can recognize each other during a match.

Rulebook generation

A rulebook generator driven by the same DSL is in progress. Both an in-session reference during play and a statically built public version are planned. A declarative DSL allows a single definition to drive multiple outputs: engine execution, natural-language rendering of legal moves, and rulebook generation.

AI research use

Longer term, opening Parlourite as a testbed for AI research is on the roadmap. Determinism, reproducibility, a unified DSL, and alignment with external benchmarks fit well with General Game Playing, reinforcement learning, and LLM agent evaluation. Serious work in this direction waits until the core is stable.

On board games

A year of this work made the diversity of board game rules genuinely striking. The structure of legal moves, informational asymmetry, resource economies, probability and preference — the range of designs produced in this field is remarkable.

Complexity and fun do not correlate. Hex and Nine Men’s Morris are still fun today. Modern mechanics are of course excellent. The field is wide, and the more time put in, the more there is to find.

A recent visit to a board game cafe with family turned into a long conversation with the owner about various titles. Thoughts about engine-building strategy kept running well after returning home. Boxes on the shelves, components spread on the table, hearing a title explained for the first time — time spent in that setting is just part of life.

The reason to keep going with Parlourite is that the project has become a vehicle for designing original games, studying strategy alone, practicing new titles, playing with relatives who live far away, and introducing games to friends. Contributing in some way to the range of what is possible with board games is enough.

Work on the UI layer begins next.