How the lab thinks

Trend bots vs Signal bots

The lab is split into two complementary groups. Trend bots (9 of them) read price and volume directly — they're momentum strategies looking for confluence between BTC's macro direction and individual altcoin behavior. Signal bots (7 of them) ignore price action almost entirely and trade off external data sources: sentiment indexes, ETF flows, funding rates, macro divergences, liquidation maps, cross-asset spreads. The two families are designed to be orthogonal — when BTC trend bots all want to go long, a funding-squeeze bot might be going short on a crowded perp, and a fear-greed bot might be sitting out entirely.

What we do before a bot goes live

The 9 trend bots went through a 5-window walk-forward validation on 60 days of hourly candles before deployment. Calibration window picks the best mode (follow vs fade) per fenêtre; test window measures out-of-sample ROI. Only setups producing positive average ROI with at least 3 out of 5 windows positive made it to paper trading.

The 7 signal bots are prototypes — they were deployed on 2026-05-20 with sensible parameters but no backtest. The lab is observing them for 1-2 weeks before keeping or pruning. Each new signal that survives gets retroactively backtested against historical data.

• Walk-forward, never curve-fit pretty backtests
• Out-of-sample only — calibration is separate from test
• Robust filter: average ROI > 0 AND ≥3/5 windows positive AND ≥5 trades total
• Paper trading first, real capital second — and even then, separately tracked
• Every entry and exit hits the database with timestamp + reason — no retrospective edits

What's under the hood

What you can trust

Every number you see on this site comes from one of three sources: live HTTP calls to the public Rankloop API, the PostgreSQL database holding closed trades, or Hyperliquid's own price feed. There are no hand-picked screenshots, no cherry-picked equity curves, no marketing P&L. If a bot is wiped, the /eth-v18 page will show its ROI in red. If it's profitable, it'll show in green. Nothing is hidden.

This is a paper trading lab. The trades you see did not move real money. The point is methodology transparency — to demonstrate that retail can build, validate, and publicly track quantitative strategies on Hyperliquid, the same way institutions do internally. If you like what the bots produce, the easiest way to support the project is to open Hyperliquid through the referral link: 4% off your fees forever, and Rankloop gets a slice of the rebates that funds further work.