Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

trade-jev

Backtest harness that tests Jev as a BUY/SELL/HOLD trader on NQ level-10 order-book data, with replay, viewer and paper-trading modes.

Projectresearch
View on GitHub ↗

Overview

trade-jev asks whether Jev can act as a short-horizon futures trader. Every 15 seconds it shows Jev the NQ level-10 order book (Databento data, 15 trading days in June 2026) and asks for BUY, SELL or HOLD, trading one contract alongside hold, random and order-imbalance baselines. Findings and published runs are in the repository.

The workflow is run once, replay many times, then view: a run calls Jev and stores answers; replay re-scores stored answers under different settings (cutoff, agreeing answers, minimum hold, stop, target, commission) without API calls; a local web app views results. A live mode runs the same harness on a live feed with paper trades and places no orders. Defaults are a 0.7 confidence cutoff and four agreeing answers. It is a research harness, not a strategy.

What the decision model decides

TypeQuestionAnswers
ChoiceGiven this order book, should the trader BUY, SELL or HOLD?BUY · SELL · HOLD

How it works

  1. Load Databento order-book data for a day.
  2. Every 15 s ask Jev for BUY, SELL or HOLD.
  3. Trade only when confidence passes the cutoff and enough answers agree.
  4. Store answers, replay under new settings and compare to baselines in the viewer.

Strengths

  • Replay re-scores stored answers without new API cost.
  • Includes baselines for fair comparison.
  • Live mode paper-trades only and places no orders.

Caveats

  • Research backtest over 15 days; not trading advice.
  • Requires paid market data and a TypeSafe key for runs.
  • Replay is approximate because settings change which positions Jev would see.

Getting started

uv sync; add TYPESAFE_API_KEY to .env; TRADE_JEV_DATA=data/sample uv run python -m trade_jev.run --days 2026-06-23 --policies hold,random,imbalance

At a glance

Type
Project
Status
research
Language
Python
Runtime
CLI + local web viewer
Has code
yes

Use cases and integrations

How curated lists describe it

Frequently asked questions

What does Jev decide?

Whether to buy, sell or hold one NQ futures contract given the current order book.

Does it place real orders?

No; backtests and the live mode are paper only.

Where are the results?

In results/FINDINGS.md and the published runs in the repository.

Related projects

  • ml-quant-trading

    PyTorch factor-research stack with 213 factors and cost-aware backtests, plus a preview typed portfolio-decision adapter.

    ProjectbetaPython
  • jev-trader

    Trade Arena proof of concept comparing trading strategies on 13 US stock funds, grown out of a pre-registered Jev study.

    ProjectprototypePython
  • jev-claim-vs-measured

    Reproducible test of viral Jev HFT claims: direction not calibrated, never under 100 ms, and no edge after trading fees.

    ServiceresearchPython
  • jev-trade-cc

    Point-in-time O'Neil momentum backtest where Jev picks entries and judges exits, with a jackknife showing no resolvable edge over rules.

    ProjectresearchPython
  • jev-typesafe-real-financial-use-cases

    Decision Lab: sixty finance demos asking Jev, OpenAI Decisions and twelve other decision models the same questions, graded.

    ProjectresearchJavaScript
  • jev-alpha-bench

    Study of whether Jev's reading of financial news predicts stock returns: strong same-day signal, no tradeable alpha.

    BenchmarkresearchGo