An old favorite, built properly.
Gomoku2D is a local-first browser Gomoku/Renju game built around one loop: play a match, then work out why it ended that way. A retro board sits on top of a Rust/WebAssembly rules core, configurable search bots, tactical hints, and Replay Analysis that shows where a loss became forced.
Live site: https://gomoku2d.byebyebryan.com/
- Start immediately against Easy, Normal, or Hard bots; no account is required.
- Choose Freestyle or Renju, including recursive Renju forbidden-move checks.
- Tune bot depth, width, scoring, and proof options through advanced settings.
- Enable hints for immediate threats, imminent threats, counter threats, and the stones behind them.
- Play on desktop or portrait mobile with dedicated touch controls.
- Keep history locally, or sign in with Google for private cloud continuity.
The aim is not the strongest possible Gomoku engine. It is a competent, inspectable opponent whose decisions connect to the same tactical language used by hints, replays, and the lab.
A finished game is more than a move list. Replay Analysis starts from the ending and walks backward through the losing side's decisions:
- Combo onset: the lethal threat that settled the result.
- Setup corridor: the forced sequence that led into that threat.
- Last escape: the final position where another move could still avoid the loss.
- Failure step: the missed response, prevention, or escape when the analyzer can classify one.
The replay can still be scrubbed normally, and any eligible frame can become a new practice game.
The Rust lab is not a separate benchmark toy. It produces the browser bot, the Replay Analysis model, and the tactical vocabulary shared by the game. Its published report exposes rankings, search telemetry, and analyzed games rather than reducing the result to a single strength number.
The Visuals guide publishes the other half of the system: pixel sprites, icons, source sheets, design tokens, and the layering rules used on the board.
Gomoku2D began as an unfinished project from nearly a decade ago and returned as a one-developer, agent-assisted production experiment. Agents made its breadth practical across implementation, analysis, review, docs, reports, and release work. The human still owns the decisions that define the project: taste, scope, domain reasoning, and technical judgment.
gomoku2d/
├── gomoku-web/ browser app: React shell, Phaser board, wasm bridge
├── gomoku-bot-lab/ Rust lab: rules, bots, analyzer, eval, CLI, wasm
├── reports/lab/ curated JSON artifacts rendered by /lab/
├── docs/ reference docs, working notes, and archives
└── scripts/ release and process-story helpers
Start in gomoku-web/ for product UI, browser state, board
rendering, and cloud continuity. Start in
gomoku-bot-lab/ for rules, bots, tactical facts, replay
analysis, evaluation, and the wasm bridge. Shared game facts originate in Rust;
the web app presents them rather than maintaining a second rules model.
gomoku-web/README.md: web app architecture, local development, build, and deploy notes.gomoku-bot-lab/README.md: Rust workspace, native commands, eval harness, and wasm bridge.docs/README.md: current product, architecture, lab, and operations references.docs/reference/product/roadmap.md: current sequencing.CHANGELOG.md: release history and intent.LICENSE: MIT license.