aaaaha0/ai-mahjong-table

How aaaaha0/ai-mahjong-table works

Mahjong Web Table combines a deterministic four-seat rules engine with a browser debug table and configurable human, random, debug, and LLM seats. The primary workflow is to configure a rule, seed, and seats in the browser; create a session; submit actions or ask a controller to choose among legal actions; then inspect snapshots, events, replay, or a FullLog. The graph preserves rule-driven play, tile/wall handling, controller integration, and inspection paths. Some runtime wiring is inferred only from module names and README capabilities because source excerpts are partial.

Main entry points: table.py Table Engine · session.py Session Manager · app.py Web API · physical_table.py Physical Operations

Components 25

Web Table

Table UI src/mahjong_ai/web/static/app.js
Web API src/mahjong_ai/web/app.py
Session Manager src/mahjong_ai/web/session.py
API Schemas src/mahjong_ai/web/schemas.py

Table Engine

Table Engine src/mahjong_ai/game/table.py
Physical Operations src/mahjong_ai/game/physical_table.py
Table State src/mahjong_ai/game/state.py
Response Windows src/mahjong_ai/game/response_window.py
Match Controller src/mahjong_ai/game/match.py

Rules and Tiles

Rule Contract src/mahjong_ai/rules/base.py
Rule Loader src/mahjong_ai/rules/loader.py
Tuidaohe Rules src/mahjong_ai/rules/tuidaohe/ruleset.py
Wall Provider src/mahjong_ai/walls/duplicate.py
Tile Model src/mahjong_ai/common/tile.py
Actions and Scores src/mahjong_ai/common/action.py

Player Controllers

Agent Contract src/mahjong_ai/agents/base.py
Agent Executor src/mahjong_ai/agents/executor.py
Random Agent src/mahjong_ai/agents/random.py
LLM Adapter src/mahjong_ai/agents/llm.py

Game Inspection

Observation Builder src/mahjong_ai/observation/builder.py
Replay src/mahjong_ai/game/replay.py
Snapshot src/mahjong_ai/game/snapshot.py
Game Events src/mahjong_ai/common/event.py

Other components

Debug Browser
LLM Endpoint

Connections 27

Debug Browser → Table UI: uses
Table UI → Web API: calls API
Web API → API Schemas: validates payloads
Web API → Session Manager: dispatches requests
Session Manager → Rule Loader: loads rules
Session Manager → Table Engine: creates and drives
Session Manager → LLM Adapter: requests decisions
Table Engine → Rule Contract: queries rule contract
Table Engine → Physical Operations: executes operations
Table Engine → Table State: updates and reads
Table Engine → Response Windows: resolves responses
Rule Loader → Tuidaohe Rules: loads plugin
Tuidaohe Rules → Rule Contract: implements
Physical Operations → Wall Provider: deals and draws
Physical Operations → Tile Model: moves tiles
Tuidaohe Rules → Actions and Scores: produces legal actions
Table Engine → Actions and Scores: accepts submitted actions
Match Controller → Table Engine: controls matches
Agent Executor → Agent Contract: runs agents
Random Agent → Agent Contract: implements
LLM Adapter → LLM Endpoint: requests action choice
Session Manager → Observation Builder: builds observations
Table Engine → Replay: provides replay
Session Manager → Replay: serializes replay
Session Manager → Snapshot: serializes snapshot
Session Manager → Game Events: serializes events
Table Engine → Game Events: records events

More Python diagrams: techwithprateek/text-to-sql · thorcollective/threat-hunting-mcp-server · titipata/yelp_dataset_challenge · tk757567/frida-patcher · tomlxxvi/pypeflow

Drawn by GitDiagram on September 28, 2026 from the repository's file tree, README and source files. Read as Markdown ·