Post
CommunityJev Practices & Patterns
Shrinking gomoku from 225 moves to 40
A gomoku harness that does tactical work locally before asking Jev, shrinking 225 possible moves to about 40 candidates per turn.
- Format
- Post
- Reading time
- 2 min
- Published by
- Community
- Author
- Víťa Vacek
- Last verified
- Sep 2026
Works forJev for Classification
Used inJev for Gaming
Highlights
- Detects wins, blocks, forks, and broken fours locally before asking Jev.
- Shrinks 225 possible moves to about 40 candidates per turn.
- Ranks candidates into tactical tiers before the model sees them.
Watch out
No win rate or comparison against asking Jev about all moves.
Questions about this entry
Short answers drawn from the linked project or page.
Effectively yes. The harness detects wins, blocks, forks, and broken fours locally, shrinks 225 moves to about 40, and tiers them before the model sees anything. Jev receives a board-level threat summary and a short continuation preview rather than the raw position.
That is the known failure mode. A reply warns that when the S and A tiers are empty and the B candidates look alike, the top Choice should not be trusted. The post itself does not address the case; it lists memory of the previous move pair as its continuity mechanism.
Possibly, but nobody measured it. The post gives no win rate and no comparison against a full-board query, so the pruning argument is architectural reasoning rather than a result. One reply calls asking over all 225 both wasteful and the wrong question.
S is a forced win or block, A is critical threats such as open fours, broken fours, and forks, B is strong attacking or defensive builds, and C is positional moves. Those definitions are the author's own, and no code or repository is linked from the post.
A short likely continuation shown for the strongest candidates. Beyond the board-level threat summary, the harness previews the leading moves with an O to X to O line so Jev sees consequences rather than having to infer them from the raw position. It is another way the harness does the inference work before the model is asked.
It carries memory. The post says the harness remembers the previous move pair and the strategic plan, which keeps a plan stable across turns instead of re-deciding from scratch. That matches the overall shape, where the local engine owns tactics and state while Jev makes the constrained choice.
From the harness's own local tactical detection, per the post, which identifies wins, blocks, forks and broken fours and drops the rest before ranking what survives into tiers. The exact cutoff is not explained and no repository is linked, so the pruning logic is not inspectable.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
Hacker News: Software Bombe and live Enigma demo where Jev ranks cribs and judges plaintext.
ag_212 pointsShow HN: Jev as the Bletchley Park Analyst: A Software Bombe on EnigmaRead the thread on Hacker NewsHacker News: Show HN for public Minecraft PvP where a diffusion LM picks moves from structured state.
2 pointsShow HN: Minecraft PvP against DiffusionGemma-as-Jev, 24 ms per decisionRead the thread on Hacker NewsHacker News: Show HN: Jev finished a Pokémon Red run, with the Hall of Fame stats on the landing page.
pancomplex272 pointsShow HN: Jev Plays Pokémon RedRead the thread on Hacker NewsHacker News: Show HN: a mouse colony where each next move is Jev or a fixed rule.
1 pointShow HN: Jev-mice – mouse colony simulation using Jev and deterministic engineRead the thread on Hacker NewsMore Jev resources like this
A video walkthrough of the control-system pattern: a request becomes structured state, Jev returns typed decisions, and code keeps the expensive model out of every loop.
Jev Practices & PatternsVideo#community#x#video
Community
A video comparison of local Laya against cloud Jev on a Tetris agent, where the open-weights model made decisions 11 times faster on a 16 GB MacBook Air.
Jev Practices & PatternsVideo#community#x#video
Community
Scott Chacon runs hosted Jev against two local decision models, laya and kev, in a Tetris match and a GitHub Settings filter, with a public test repo.
Jev Practices & PatternsVideo#community#video#open-models
Community
A Japanese experiment that builds a rules-based gomoku harness and replays a full Jev-versus-Jev game log to measure real per-move latency.
Jev Practices & PatternsArticle#community#article#games
Community
A pointer to Diogo Almeida's independent working note on building a Jev harness for coding agents, including why harnesses already make these calls badly.
Jev Practices & PatternsPost#community#x#coding-agent
Community
A ViZDoom agent with two Jev decision channels, navigation at 5 decisions per second and combat at 12, finishing a test run with 18 kills.
Jev CookbooksPost#community#x#video
Community
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.