Disclosure: Deploy Finance publishes this article and runs the Income: Funding Rates and Superstar agents it mentions.
Not financial advice. Crypto, DeFi and perpetual-futures strategies can lose money, including your whole deposit. Yields are variable, and past or backtested results do not predict future returns. Read the risk disclosure.
Seventy days after Superstar started trading real capital, every wallet running the agent is in profit. That holds for wallets that joined on day one and for those that joined weeks later. It is the milestone we set out to reach by the end of the second phase of our rollout, and we reached it with the agent still using only a fraction of the capital it will run at full speed.
At a glance
- 100% of agent wallets profitable at the end of the second rollout phase, across every start date.
- Steady, low-risk gains. A representative $108,000 agent wallet made +$1,561 net, won 97% of its 308 trades, had no losing month, and its worst drawdown was 1.2% of the balance. Its Sharpe ratio is 2.70.
- A strong edge on the capital at work. The money the agent actually had in positions earned about 108% (annualised if equated to 100% balance usage) a year.
- Returns climbing with each phase. Each time we loosened the agent's limits, returns went up: 5.7% annualized in the first 30 days, 7.3% in the next 30, and 14.2% so far in the third phase. Over the same time, average use of the balance grew from 2% to 21%.
- Next 30 days: use of the balance rises to 100%, perp agents start running on top of spot-trading sub-agents in the same wallet, and HIP-4 prediction markets enter live beta.
Why we rolled out in phases
Moving an AI trading agent from paper trading to real capital is done in steps, not all at once. Each step raises the share of the balance the agent may use, but only after it has behaved as expected at the previous size.
- Capital protection comes first. Our users trust Superstar with real money. Starting small meant the agent could prove itself in live markets while the amount at risk stayed small.
- Live markets teach what a backtest cannot. Fills, timing and position management behave differently with real orders. Each phase let us tune the agent in real time, while any issue cost little.
- Size grows with evidence. Every increase in the agent's limits was earned by the results of the phase before it.
That is why the agent spent its first two months using only part of each wallet. It was the design, not a limitation.
What the first 70 days delivered
Our representative wallet finished its first 70 days at +$1,561 net. It made money in every month, and its worst drawdown never went past 1.2% of the balance.
| Measure | Result, Aug 12 – Oct 20, 2026 |
|---|---|
| Net PnL | +$1,561 on a $108,235 balance |
| Trades | 308 round trips, 97.1% of them profitable |
| Profit factor | 8.56 (dollars won per dollar lost) |
| Sharpe ratio | 2.70 (Sortino 3.63) |
| Worst drawdown | 1.2% of the balance |
| Net PnL by month | August +$505 (from Aug 12) · September +$320 · October +$737 (to Oct 20) |
The agent traded BTC, HYPE, ETH and ZEC perpetuals. It added HIP-3 stock perpetuals (NVDA and SPCX) in mid-September and spot BTC in October. October, the first month with more room to work, is already its best month.
The utilisation story: returns rise as the agent is given room
Every time we loosened the agent's limits, returns went up. Average use of the balance grew tenfold over the rollout, and the annualized return more than doubled.
| Phase | Dates | Average use of balance | Peak use | Net PnL | Annualized return |
|---|---|---|---|---|---|
| Phase 1 | Aug 12 – Sep 11 | 2.0% | 18% | +$504 | 5.7% |
| Phase 2 | Sep 11 – Oct 11 | 7.7% | 37% | +$653 | 7.3% |
| Phase 3 (first 9 days) | Oct 11 – Oct 20 | 20.5% | 35% | +$404 | 14.2% |
The headline return is modest because most of the balance was deliberately left idle. The money the agent did put to work tells the real story: over the whole period it earned about 108% a year. In the latest phase, at larger sizes, it earned about 70% a year. That is the edge the rest of the balance gains access to as utilisation rises.
Live vs backtest, compared fairly
Our published backtest returned 171.5% with a worst drawdown of 13.6%. Live results look lower at first glance, but the two are not on the same footing: the backtest put 100% of the balance to work from day one, while the live agent averaged 7%. Scaled to the same 100%, the live record lines up closely.
| Backtest | Live, scaled to 100% of balance | |
|---|---|---|
| Share of balance in positions | 100% | 100% (scaled from 7% actual) |
| Return | 171.5% total | About 108% a year |
| Worst drawdown | 13.6% | 14.1% |
| Profitable trades | 52.3% | 97.1% |
| Market | BTC up to about $124k, then down to about $63k | BTC up about 40%, then steady near its highs |
The risk profile is almost identical, and the live agent wins far more of its trades. The markets also differed. The backtest ran through a sharp fall, where short trades did much of the work. Live, BTC rose and then held, so the agent stayed long almost all the time. Superstar trades both directions, and can switch when the market turns.
What comes next: the next 30 days
The next phase takes Superstar from part of the balance to all of it, and adds the capabilities that let it do more with that capital.
- Utilisation rises to 100%. Over the coming 30 days, use of the balance steps up from today's peak of about a third to the full balance. This brings live trading onto the same footing as the backtest.
- Perp agents on top of spot sub-agents. Perp-trading agents will run alongside spot-trading sub-agents inside the same wallet portfolio. Spot positions such as BTC can be held through uptrends and also serve as collateral for perp trades, opening the door to more sophisticated strategies.
- Stock perpetuals at work. HIP-3 markets already let the agent trade stocks such as NVDA. As equity markets move, these pairs add a new source of return beyond crypto.
- HIP-4 prediction markets in beta. Live beta testing of HIP-4 prediction markets begins on selected wallets.
The road to full speed
The first 70 days answered the question that matters most before scaling up: does the agent make money with real capital, for every user, whenever they join? It does. Every wallet is in profit, returns have risen with each phase, and the capital the agent puts to work is earning at a rate that lines up with our backtest.
So far Superstar has shown what it can do with a fraction of the balance. The next 30 days give it the whole balance, a spot portfolio to build on, and new markets to trade. We are excited to see what it does with the room.
About these figures
Performance figures come from one representative agent wallet, from Aug 12 to Oct 20, 2026, net of all costs. The share of profitable wallets covers every Superstar wallet. Figures scaled to 100% of the balance replay the same trades at a larger size; they show what the agent's record implies, not a forecast. Past performance does not guarantee future results, and nothing here is investment advice.