How it works
Superstar runs on a fixed schedule.
Every four hours it takes a fresh read of current market data and makes one decision: open a long, open a short, or stay out of the market.
It works across spot, perpetuals, and HIP-3 markets on Hyperliquid, taking a position wherever the read is strongest, and takes outcome-market positions selectively when there is a clear winner. The market read below covers the signals it weighs on the leveraged markets.
It carries no view from prior cycles and does not wait for you to confirm anything.
In the backtest it declined to trade on 1,350 of 2,298 reviews.
The decision loop
Every four hours, the agent walks the same four steps on the current market state. Each step has a clear output that feeds the next one.
| Step | What it does | Output |
|---|---|---|
| 1. Read the last four hours | For each input, lay out where the reading stood 3 hours ago, 2 hours ago, 1 hour ago, and now. | Time-series of the last four hourly prints |
| 2. Synthesize a picture | Map price and OI quadrant, order-book lean, liquidation clusters, funding and volume trends. Missing feeds are marked and weighted down. | A composite market state across the four quadrants |
| 3. Argue both sides | Write an explicit bull case and bear case, each with concrete points and the exact level that proves the idea wrong. | Two opposing theses with hard invalidation levels |
| 4. Choose a direction | Long, short, or no trade. Carries a rough probability, entry plan, invalidation level, and what would justify adding. | A decision plus the plan to act on it |
When the cases are close or the data is thin, the verdict is no trade.
Market inputs the agent weighs
The agent reads the market across four lenses. Each lens contributes a handful of measurable inputs.
Price context
| Input | What it captures |
|---|---|
| Price 1h close | The anchor for the current read |
| 1h high | Where price was pushed and rejected in the last hour |
Positioning and leverage
| Input | What it captures |
|---|---|
| OI 1h close | How much leveraged money is in the market |
| OI delta last hour | Whether positioning is building or unwinding |
| Realized funding | Which side has paid funding recently |
| Indicative funding | Which side will pay on the next settlement |
Forced buying and selling
| Input | What it captures |
|---|---|
| Long liquidation pool | Where crowded longs would be force-closed |
| Short liquidation pool | Where crowded shorts would be force-closed |
| Long liquidations 1h realized | Where forced long-closes already fired |
| Short liquidations 1h realized | Where forced short-closes already fired |
Order flow and pressure
| Input | What it captures |
|---|---|
| Buy volume 1h | Whether real buying is behind a move up |
| Sell volume 1h | Whether real selling is behind a move down |
| Bid/ask ratio 0 to 2% | Order-book micro-pressure |
| Daily CVD | Cumulative volume delta over the trading day |
Whales versus retail
| Input | What it captures |
|---|---|
| Whale-retail delta | Where large accounts differ from the crowd |
| Global retail long/short | The crowd direction across all users |
| True retail long/short | The crowd direction excluding market makers |
| Top trader long/short | Where the largest accounts sit |
The agent uses these inputs as cross-market confirmation across exchanges, timeframes, and options conditions.
Risk management
Risk is part of the decision to trade. If the risk does not fit, the trade does not open.
| Rule | How it works |
|---|---|
| Loss limit set before entry | The maximum loss is decided before the trade opens and is never widened afterward |
| Position size comes from the limit | Sizing is mechanical. Size follows the loss limit, not the conviction in the idea |
| Stop only moves to protect capital | The stop trails behind price as a trade works. It never widens to give the trade more room |
There are no revenge trades after a loss. Most losing trades end the same quiet way, at their stop.