User repository on GitHub
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。
- Forks
- 0
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
A gomoku lab that runs nine 15x15 games at once with two Jev players, holding the rules fixed while you vary how much tactical detail reaches the model.
User repository on GitHub
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。
Repo stats from the GitHub API, cached Oct 2026.Homepage
Watch out
The matchup is not unassisted. Candidates are heuristic-filtered to your own top eight, the opponent's top four plus all immediate wins and direct blocks, so the model chooses from a shortlist rather than every empty point. Option probabilities are explicitly not win rate, there is no Score or Noul toggle and no win-rate configuration, and no chain of thought is collected. The README states that nine-game outcomes depend on opening, side, model version and sampling, so a result cannot on its own show that one input scheme is better. The short-term lookahead is computed by local rule code, not by the model, and the four-direction analysis only looks at consecutive stones. Live play bills your own OpenRouter account, and pausing does not cancel already-submitted upstream requests, which may still be charged. Copy reviewed against the repository README on 2026-10-04, not from running it.
Original write-ups on what this entry is for.
Posts from builders shipping with Jev right now.