# 0xArchive open-sources a quant research skill for coding agents

> Original title: Reverse engineer any trader. Backtest anything.

- Source: X search: AI open-source releases
- Published: 2026-10-25T18:52:59.000Z
- Bad Influence: https://badinfluence.news/items/dwmki91humoogemjnkpxit32e
- Original: https://x.com/0xArchiveIO/status/2114430034832433278

## Summary

0xArchive has open-sourced a quant research skill that lets coding agents backtest trading strategies on its Hyperliquid data. The skill compares each strategy against three fake baselines and logs every trial. Its 7-day momentum example reports a Sharpe of 2.43 in-sample and 1.81 out of sample, with a deflated Sharpe of 0.93 that falls below its 0.95 bar, so it is labeled as holding out of sample rather than an edge.

## Why it matters

The post shows how the skill tests a strategy against randomized baselines and logs every trial, which makes its verdict easier to check than a single backtest.

## Post

Reverse engineer any trader. Backtest anything.

We just open-sourced a quant research skill for coding agents. It runs on 0xArchive's Hyperliquid data. Hand your agent a wallet, an idea, a Pine script or a paper, and get a verdict anyone can rerun.

http://github.com/0xArchiveIO/0xarchive-quant

Paste a wallet and the agent will:

- pull every fill that changed a position
- profile the style: which markets, how long it holds, when it opens
- mirror the book in a Rust backtester with fees and funding
- write candidate rules and score how much of the book each one reproduces

Then it tests skill against luck. It puts the book up against three fakes: the same trades flipped, random positions in the same markets, and the trader's own positions moved to random earlier times. If the real book beats all three, the timing was worth something.

We checked that it finds the right rules on a book where we knew the answer: 297 position changes made by one of our example strategies. The real rules matched the trader's side in 100% of held hours, with 0.97 correlation in daily returns. An unrelated strategy got 2.5% and -0.02.

We ran our own idea through it too: 7-day momentum on the top 25 perps over 16 months. Sharpe 2.43 and +71.4% after fees and funding, none of the 20 random or stale runs matched it, and Sharpe 1.81 out of sample.

It still isn't called an edge. We tried 4 settings, and over 4 tries the deflated Sharpe is 0.93, under the 0.95 bar. So the verdict is "holds out of sample", rung 4 of 6.

Every try is logged, and the last months of data stay sealed until you commit one strategy, so the tenth tweak can't pass as the first.

The engine, for any agent:
pip install "0xarchive-backtest[curated]"

Claude Code:
claude plugin marketplace add 0xArchiveIO/0xarchive-quant
claude plugin install 0xarchive@0xarchive-quant

Codex:
git clone https://github.com/0xArchiveIO/0xarchive-quant.git
mkdir -p .agents/skills && cp -R 0xarchive-quant/skills/backtest .agents/skills/0xarchive-backtest

Then ask with /0xarchive:0xarchive-backtest in Claude Code or $0xarchive-backtest in Codex.

A free API key covers the last 30 days.

![](https://badinfluence.news/api/img-proxy?u=https%3A%2F%2Fpbs.twimg.com%2Fmedia%2FHVf1VG_b0AA7RHa.jpg&mode=thumb&exp=1793404800&sig=d58ffe94aeb6e7f5)