Can this strategy actually make money on Polymarket?

No proven profitability is claimed: entry needs fee-inclusive edge over a shrunk score, parameters are adjustable research settings, and the demo on synthetic data says nothing about strategy returns.

What prevents stale prices or a mid-round restart from faking an edge?

It verifies the 300-second market, reads the settlement source per round, requires exact opening-price evidence, skips unknowable mid-round starts, and discards results when round, price, or books go stale.

Why does the dashboard show a lean but the trader waits?

The dashboard always shows raw Up and Down scores with the lean, while entry, holding, spread, edge, and re-entry rules decide separately whether the paper trader buys, sells, or waits.

What breaks if I set live mode or expose the server publicly?

Live mode still paper-trades and needs explicit data, model, key, and ledger settings; keep it on localhost, set control, origin, and memory options for server use, and never touch real collateral.

What do I need to run the decision lab demo?

Bun, verified locally with 1.2.6. Run bun install --frozen-lockfile then bun run demo, and open http://127.0.0.1:3000. The demo forces deterministic synthetic prices, simulated order books and a mock model, so it never calls Jev and says nothing about strategy returns.

How often does the lab ask Jev, and when does it stop?

By default the model is called every 6 seconds (POLL_SCHEDULE 300:6000) and not at all in the last 30 seconds of a round (MODEL_STOP_SECONDS_LEFT), after which open positions are held to official settlement. Scheduling is start-to-start and only one request is in flight.

How are Jev's scores turned into buy and sell decisions?

Jev's class scores are not calibrated win rates, so they can be shrunk toward 0.5 with direction score = 0.5 + SCORE_WEIGHT * (raw score - 0.5). An entry needs an edge of at least MIN_EDGE over the fee-inclusive cost; the default shrink weight is 0.5 and the default minimum edge is 0.05.

CommunityJev Tools & CLIs

Jev Poly Crypto Demo

Research dashboard for Polymarket BTC five-minute Up/Down rounds that feeds market state to Jev class scores and paper-trades the edge.

See highlights

Open github.com

Stars: 38
Language: TypeScript
Published by: Community
Author: frankda
Last verified: Oct 2026

frankda/jev-poly-crypto-demo

User repository on GitHub

View on GitHub

Forks: 13
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.Homepage

About this project

A research tool for Polymarket's BTC 5-minute Up/Down markets, following the "market state → Jev class scores → independent execution logic → live dashboard" structure of jarrodwatts/jev-trader. This version implements a live market-data adapter, Jev integration, paper trading and a dashboard. It does not connect a wallet or submit real orders. TRADINGMODE=live is rejected at startup. Requires Bun (verified locally with 1.2.6).

Quoted from the project's README. Read the full README.

Tagscommunity jevbest catalog-import

Works forJev for Classification

Used inJev for Software & DevTools

Languages: TypeScript

Highlights

Derives round ids from UTC time, verifies the 300-second market, maps outcomes by label, and resolves against the market's own settlement source.
Asks Jev whether the round ends at or above the opening price using books, time left, and optional perpetual features that never settle.
Enters only on fee-inclusive edge with one open position, rechecks bids for exits, holds to official settlement, and never guesses unconfirmed outcomes.
Runs a synthetic demo without model calls plus a live-data paper mode with staleness discards, single-flight polling, and ledger recovery.

Watch out

Paper trading only with no wallet and live order mode rejected; class scores are not calibrated win rates and simulated fills ignore queue and competition. Do not expose the local dashboard as-is. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

jev-recall

Python memory filter that judges every stored memory against the current request with one calibrated yes/no per memory in a single Jev request, keeping everything above a threshold.