jevymarket
Experimental Polymarket bot: a web-search model writes an evidence brief, Jev prices P(YES), and code trades only on a configured edge.
Overview
jevymarket is a Polymarket trading bot whose pricing oracle is Jev, called through OpenRouter's Decisions API. Because Jev has no browsing and a training cutoff, it rightly refuses to price news-driven markets on its own, so the bot pairs it with a cheap researcher: a generative model with web search writes a dated, sourced evidence brief into the state, with odds sites excluded so Jev never sees market prices as evidence. The state also includes pre-computed days to resolution, since Jev is weak at date arithmetic.
One Jev call asks three questions: the probability the market resolves YES, whether the state holds enough current information to form an informed estimate, and how objective the resolution criteria are on a 0 to 4 scale. Pure evaluation code skips unless the market is answerable and clear, only buys within a price band, requires the probability minus the ask to exceed a minimum edge, and sizes with capped fractional Kelly. Execution places GTC limit orders only if hard caps on per-trade size, open exposure, trades per run and one position per market pass. Everything is logged to SQLite. The README warns it places real orders with real money.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Will this market resolve YES? | — |
| Yes / no | Is there enough current, relevant information in the state to form a well-informed estimate? | — |
| Score | How objective are the resolution criteria? | 0 · 1 · 2 · 3 · 4 |
How it works
- Scan open markets by volume and filter by liquidity, spread and timing
- Research model writes a sourced evidence brief (cached)
- Jev answers resolves_yes, answerable and clarity in one call
- Evaluate edge versus the ask, price band and Kelly sizing
- Execute limit orders within hard caps and log everything
Strengths
- Separates evidence gathering from calibrated judgment
- Multiple hard caps on exposure
- Answerability gate avoids trading on thin information
- Full audit log
Caveats
- Trades real money; experimental
- Edge depends on Jev's calibration in prediction markets, which is unproven
- Relies on a researcher model's evidence quality
At a glance
- Type
- Project
- Status
- prototype
- Language
- Python
- Runtime
- CLI bot
- Has code
- yes
- Has tests
- yes
Use cases and integrations
How curated lists describe it
Polymarket trading bot driven by Jev (TypeSafe AI) via OpenRouter.
hellogumbo/awesome-jev
Frequently asked questions
What does Jev decide?
The probability a market resolves YES, whether there is enough information, and how clear the resolution criteria are.
Why is a second model needed?
Jev cannot browse, so a web-search model supplies a dated evidence brief.
Is it safe to run?
It places real orders; the README asks you to read the disclaimer and security notes first.
Related projects
Local China A-share technical analysis workbench where Jev assesses eight indicator dimensions and returns Buy, Hold or Sell with evidence.
ProjectbetaPython, TypeScriptStock terminal that gathers live evidence via Valyu and asks Jev for the probability a stock trades higher in 30 days.
ProjectprototypeTypeScriptTerminal agent for Polymarket 5-minute BTC markets that asks Jev up or down and buys once only above 0.90 confidence; dry-run default.
ProjectprototypeTypeScriptPolymarket BTC 5-minute up/down agent: code prices with a volatility model, Jev scores regime and vetoes anomalies; paper by default.
ProjectprototypePythonPaper-only research tool for Polymarket BTC 5-minute Up/Down markets: Jev scores the direction, separate code decides entries.
ProjectresearchTypeScriptExperiment that builds past-only Hong Kong market state and asks Jev for an up, flat or down forecast, with a standalone HTML report.
ProjectresearchPython