How Super Perps works
Super Perps runs on a fixed schedule. Every four hours it takes a fresh look at the market and makes one decision: open a long, open a short, or do nothing. It does not carry over yesterday's opinion, and it does not wait for you to confirm anything.
The loop
Each cycle starts from scratch with current data. The agent decides, and if it trades, it sizes and protects the position before the order goes in. Then it waits for the next cycle and repeats.
Running on a schedule matters more than running constantly. The agent is not trying to react faster than a human. It is trying to remove the need to react at all. In the backtest it completed 2,298 reviews and chose not to trade on 1,350 of them, so most of the time the right move was to wait. When it did trade, its winners outweighed its losers: about $1.27 came back for every $1 the losing trades gave up (366 winning trades to 322 losing ones).
How a decision gets made
Before it picks a side, Super Perps argues with itself. Every review runs the same four steps.
1. Read the last four hours, hour by hour
For every input on the signals page, the agent lays out where it stood three hours ago, two hours ago, one hour ago, and now. The trail matters more than the latest value. A funding rate that is high and falling means something different from one that is high and still climbing, and a buy-volume spike that is already fading is a warning a single snapshot would miss.
2. Synthesize the readings into a picture
It pulls the readings together into a few structural questions. The first is the price and open-interest quadrant: what kind of move is this?
Real bid behind the move. The most trustworthy kind of rally.
A squeeze, not fresh demand. Weaker fuel that can fade fast.
Fresh selling pressure entering as price falls.
Positions unwinding, not aggressive new selling.
Price direction plus open-interest direction tells the agent what kind of move it is looking at.
Alongside that it maps where the order book leans, where liquidation clusters sit above and below the price, and which way funding and volume are trending. When a feed is missing, and some venues do not publish every positioning metric, the agent marks that input unavailable and leans on it less rather than guessing.
3. Argue both sides
The agent then writes an explicit bull case and bear case, each as a list of concrete points: the higher-timeframe structure, what positioning implies, whether buying or selling is exhausting, and the exact level that would prove the idea wrong. Forcing out the strongest argument against its own lean is what keeps it from talking itself into a weak trade.
Doing nothing is a valid move. In the backtest, 1,350 of 2,298 reviews ended in no trade.
4. Commit to a verdict
Finally it commits: long, short, or no trade. The verdict is more than a direction. It carries a rough probability, a plan for where to enter (chase the breakout, or wait for a pullback into a retest zone), the level that invalidates the idea, and what would make it add. If neither case is strong enough, or the data is too thin to see the market clearly, the verdict is no trade. That is most of the time, and it is not a failure state. Sitting out a coin flip protects capital for the setups worth taking.
What happens after it trades
Once a position is open, the agent stops asking whether to be in the trade and starts managing the one it has. It holds the loss limit it set at entry, and as the trade moves into profit it protects more of the gain. The risk management page covers exactly how that works.
Read next
- Signals and edge: the market inputs behind each decision.
- Risk management: sizing, stops, and profit protection.
- Backtest performance: how the loop held up across a full cycle.