BTC 5m Decision Lab

How we reviewed this: we ran its bun run check, then started the server offline in demo mode and tried to start it in live mode to see whether the refusal is real. We made no Jev calls and connected no wallet.

What it is

A decision lab for Polymarket’s BTC 5-minute Up/Down markets, following the “market state → Jev class scores → independent execution logic → live dashboard” structure of jev-trader. Market data comes in, Jev scores the direction, separate code decides whether that is an edge worth taking, and a dashboard shows both.

The refusal is real

The README says: “It does not connect a wallet or submit real orders. TRADING_MODE=live is rejected at startup.”

We tried it:

$ TRADING_MODE=live bun run src/index.ts
error: Only TRADING_MODE=paper is implemented; real orders are not supported
exit 1

And in the source it is a throw during config validation, not a warning:

if ((env.TRADING_MODE ?? "paper") !== "paper")
  throw new Error("Only TRADING_MODE=paper is implemented; real orders are not supported");

Note the shape: !== "paper", so any other value is refused, not a blocklist of "live". For a repository that a stranger might clone while thinking about money, that is the right way round — the safe value is the only value, and the unsafe one is not a flag you can find.

Demo mode started cleanly with no network and no keys:

JEV / POLY → http://127.0.0.1:3000 · PAPER · demo data · mock model · sqlite ledger

The banner names all four choices back to you, which is a good habit for anything that could be pointed at real data by accident.

What it says about its own numbers

The README does not claim a strategy works:

The demo uses deterministic synthetic prices, simulated order books and a mock model. It exercises entries, 5-minute round changes, settlement and ledger recovery end to end, but says nothing about strategy returns.

The default shrink weight is 0.5… There is no minimum direction score by default… These are adjustable research parameters with no proven profitability.

That is the correct claim for a tool of this kind, and it is unusual in this corner of the ecosystem. The dashboard also separates the raw Up/Down scores from the paper trading action, so the model’s lean and the code’s decision are not presented as one thing.

The scheduling is the engineering

The interesting work is not the Jev call, it is when not to make it. From the README, and reflected in the tests: the model is called every six seconds by default, not at all in the last 30 seconds of a round (positions open then are held to official settlement), scheduling is start-to-start so a slow request does not stretch the interval, only one request is ever in flight, timeouts skip missed slots instead of piling up, “a tick never spills into the next round”, and failures back off exponentially.

Its tests read like that list — “a slow last-minute tick never spills past the next round’s opening”, “errors back off from completion and retain the one-minute ceiling”, “client state rebuilt from patches equals the server’s full state across cycles and a round change”. 66 tests pass across 10 files, and tsc --noEmit is clean.

The licence

There is no licence file. For a repository about trading, that is the thing to fix first: without one, nobody can legally fork it, and “it’s on GitHub” is not a grant. Twenty-nine stars.

Verdict

Better built than its subject matter usually attracts. The safety property is real and enforced in the one place that matters, the demo runs offline so you can see the whole loop without an account, and the author declines to claim the strategy makes money.

Treat it as what it says it is: a lab for watching a decision model score a market in real time, with paper money. Add a licence before you build on it.

For the project whose structure it follows, see jev-trader; for another financial decision tool, Prism Liquidity Agent.

See how it compares with other tools in Best Jev tools, tested hands-on.

Review updated Sep 22, 2026. Numbers quoted from the project are its author's own; we don't publish our own measurements of Jev.