Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

jev-gomoku (XieChengYuan)

Nine-board gomoku workbench comparing how two Jev players respond to five different input representations of the board.

Projectresearch
View on GitHub ↗Live demo ↗

Overview

Yishun (弈瞬) is a Chinese-language gomoku workbench that runs nine 15x15 games at once between two Jev players who alternate colours, asking how the same Jev plays when given different information. Each move is one Choice question selecting the best point from candidates generated by the same rule (top heuristic moves for each side plus all immediate wins and blocks). Five input schemes are compared: board only; board plus tactical facts; plus short-term lookahead computed by local code; board plus coordinate lists; and board plus four-direction line counts.

A live dashboard shows state, questions, answers, usage, model version and candidate probabilities for every move. The online site offers a free replay of 144 saved real decisions and a live mode using the visitor's own OpenRouter key, kept only in page memory. The README is explicit that candidates are pre-filtered so this is not an unaided strength test, Choice probabilities are not win rates, and nine games cannot prove one input is better.

What the decision model decides

TypeQuestionAnswers
ChoiceWhich candidate point is the best move?—

How it works

  1. Generate candidate moves with the same rule for both players
  2. Build state using the selected input scheme
  3. Ask Jev one Choice per move
  4. Record requests, probabilities and outcomes across nine boards

Strengths

  • Controlled comparison of input representations
  • Full per-move transparency
  • Free replay with no API calls

Caveats

  • Candidates are pre-filtered
  • Nine games are too few to draw conclusions
  • Chinese documentation

Getting started

Try https://xiechengyuan.github.io/jev-gomoku/ or run locally with Node 22+: `npm ci`, set OPENROUTER_API_KEY in .env, `npm run dev`.

At a glance

Type
Project
Status
research
Language
TypeScript
Runtime
Browser app (GitHub Pages) with optional local server
Has code
yes
Has tests
yes

Use cases and integrations

How curated lists describe it

  • A nine-board, 15×15 Gomoku workbench comparing how two Jev players respond to different input representations.

    logicrw/awesome-jev-projects

  • 弈瞬：双 Jev 五子棋九宫格输入实验台，逐手查看模型决策，支持真实对局回放与实时对战。

    hellogumbo/awesome-jev

Frequently asked questions

What is being tested?

How different board representations change Jev's move choices.

Can I try it without a key?

Yes, the replay mode plays back 144 saved real decisions.

Do the results prove one input is best?

No, the README says nine games are not enough.

Related projects

  • jev-behavior-study

    Independent field guide to Jev 1.13.0 behaviour across 11,621 requests: framing sensitivity, Snake and 3D city driving experiments.

    ProjectresearchPython, TypeScript
  • jev-vs-decisions-rpg

    RPG experiment pitting Jev against OpenAI's Decisions API as NPCs, comparing wins, per-move accuracy, calibration and cost.

    ServiceresearchPython
  • Typesafe_chess_eval

    Short evaluation finding that Jev plays chess poorly, often calling unbalanced positions draws.

    Projectresearch
  • jev-demos

    Small demo collection testing Jev's spatial reasoning with maze single-step choices versus multi-step lookahead.

    AppprototypePython
  • PZ_Optimization

    Project Zomboid performance patches whose benchmark harness computes numbers in code and, per a list, asks Jev only for the verdict.

    BenchmarkbetaJava
  • Jev Playground

    Benchmarks Jev against Luna, Haiku, Gemini and Astra at choosing legal moves in tic-tac-toe and Connect Four.

    BenchmarkprototypeTypeScript