tateti-jev
Tic-tac-toe teaching project comparing Jev Choice, Score and Noul designs across three levels of state detail.
Overview
Ta-te-ti x Jev is a Spanish-language tic-tac-toe project for learning how to design decisions for Jev, following the rule that code calculates and Jev decides. It compares the three primitives, Choice, Score and Noul, and three levels of state: no facts, win/block hints, and complete per-square facts. A shared engine handles rules, minimax, per-square facts, baseline controls (random, minimax, fixed rule) and question construction.
A runner makes measurement calls to Jev 1.13 through OpenRouter with a spending cap, and a Cloudflare Pages site hosts the page, with the OpenRouter key kept as a server-side secret. The README reports results from 4,110 calls costing US$0.21: with full facts all three primitives played optimally, the same as a fixed rule with no AI, while with no facts Jev was close to random. The lesson drawn is that the state you send matters more than the primitive.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which square should be played next?(inferred) | — |
| Score | How good is this square as a move?(inferred) | — |
| Yes / no | Is this square a good move?(inferred) | — |
How it works
- Engine computes legal moves and, depending on level, facts per square.
- Build Choice, Score or Noul questions from those facts.
- Send to Jev via OpenRouter and play the chosen move.
- Compare outcomes against random, minimax and fixed-rule controls.
Strengths
- Cheap, controlled experiment on question and state design.
- Baselines make the result easy to interpret.
- Public page to try it.
Caveats
- Toy domain; findings may not transfer directly.
- Documentation is in Spanish.
Getting started
Open https://tateti-jev.pages.dev, or run correr.mjs with an OpenRouter key and a --limite spending cap.At a glance
- Type
- Project
- Status
- research
- Language
- JavaScript
- Runtime
- browser + Cloudflare Pages Function
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Ta-te-ti para aprender a diseñar decisiones con Jev (TypeSafe): Choice vs Score…
wh000wh000/awesome-jev-live
Frequently asked questions
What did the experiment find?
With complete facts all three primitives played optimally; with no facts Jev was close to random, per the README.
How much did it cost?
The README reports 4,110 calls for US$0.21.
How is Jev accessed?
Through OpenRouter's decisions endpoint using typesafe/jev-1.13.
Related projects
Independent field guide to Jev 1.13.0 behaviour across 11,621 requests: framing sensitivity, Snake and 3D city driving experiments.
ProjectresearchPython, TypeScriptFifteen live arcade games where Jev and LLMs play each other, led by WikiRace in which Jev picks only real links.
ProjectbetaTypeScriptOne-page web app where Jev plays chess against OpenRouter LLMs, Stockfish or you, showing the moves it weighed and their probabilities.
AppbetaJavaScriptPac-Man where Jev or other System One models choose each move at junctions, with live odds, fallbacks and a leaderboard.
ProjectbetaTypeScriptBrowser chess lab where Jev, LLMs and people play, each Jev turn being one Choice over the legal moves, with Stockfish analysis.
ProjectbetaPythonNiceGUI desktop chess app for evaluating Jev's move choices against Stockfish, random players or itself, with PGN export.
ProjectprototypePython