TL;DR: Paper trading runs your bot forward on live prices with virtual money, so it tests the one thing a backtest cannot: whether the strategy behaves the same on data that did not exist when you built it. On Trigr, a paper agent runs the same signal pipeline as a live Hyperliquid agent but fills at the current Hyperliquid price with fees deducted and no slippage or funding modeled. Use it to verify behavior, not to forecast net profit.
Why forward-test at all if the backtest looks good?
A backtest is a claim about the past. However carefully it is built, you chose the strategy after seeing that past, often after trying many variants. That selection process is what the Deflated Sharpe Ratio and the Probability of Backtest Overfitting try to correct for, and it is why reading a backtest critically matters.
Forward testing removes the selection problem for the period it covers. Nothing to the right of the start date could have influenced your design. Paper trading is the lowest-cost way to get that clean record, because the only thing at stake is time.
It also catches a different class of error: plumbing. A signal that looked fine on historical bars may fire at a surprising moment live, depend on a data source that updates late, or interact badly with other strategies sharing the same capital.
What does a Trigr paper agent actually do?
A paper agent is one of three venue choices for a Trigr agent, alongside Hyperliquid and Propr. It has up to 52 strategy slots, one per market, each with its own allocation and RISK node. New entries size from current simulated account equity, which changes with gains and losses. Initial capital funds the account and remains a performance seed.
When a strategy emits a signal, the paper agent opens a virtual position at the current Hyperliquid price and tracks P&L in a per-agent virtual portfolio, which starts with $10,000 of virtual capital. Each fill deducts Hyperliquid's base-tier taker fee and the builder fee; there is no slippage and no funding accrual, and leveraged positions carry an approximate liquidation level like the backtest's. Take-profits, stop-losses, trailing stops and time stops defined in the strategy's RISK node manage the exit, just as they would live.
On the agent's chart, a frontier marker separates the backtested segment from the forward record. Everything right of the marker is genuine forward performance. Paper history and live history are kept separate, so a paper record is never blended with real fills.
Separately from your own agents, every running strategy on Trigr keeps a canonical paper track: one system-run forward record, sized at a fixed slice so curves are comparable across strategies. That is the live-performance figure the strategy marketplace shows next to each verified backtest.
What can paper trading prove, and what can't it?
| Question | Paper agent answers it? | Why |
|---|---|---|
| Do signals fire when the backtest says they should? | Yes | Same signal pipeline as live |
| Is trade frequency in line with the backtest? | Yes | Real calendar time |
| Do multiple slots fit together without over-exposure? | Yes | Same allocation and risk panel |
| What will my net return be? | No | No slippage or funding in fills |
| Will my orders fill at these prices in size? | No | No order book or market impact |
| Will the edge persist? | Partly | Only for the period observed |
The cost gap is the big one. Paper fills already deduct a base-tier taker fee and Trigr's builder fee of up to 0.05% per trade, but live trading adds slippage on every fill, and positions pay or receive funding every hour. A strategy trading many times a day can see a large share of its paper edge disappear to those costs. The honest comparison for a paper record is therefore a backtest run with slippage and funding switched on, since Trigr's backtests are gross of those by default.
How long should you paper trade?
Think in trades, not days. A handful of trades tells you almost nothing about expectancy; it mostly tells you whether the plumbing works.
Some practical guidance:
- High-frequency strategies (5m to 30m bars): a few weeks often yields dozens of trades, enough to compare win rate and average trade against the backtest.
- Intraday to swing (1H to 4H): expect one to three months for a comparable sample.
- Daily and weekly strategies: paper trading mainly confirms behavior; statistical confirmation would take many months, so lean harder on backtest quality (point-in-time data, DSR, PBO) instead.
A useful habit is to write down, before starting, what result would make you abandon the strategy: for example, a drawdown deeper than the backtest's worst, or a win rate far below it after a set number of trades. Deciding the rule in advance keeps you from rationalizing a poor forward record after the fact.
Also note the market regime while you test. A trend-following strategy paper-traded through a flat month will look worse than its backtest for reasons that say little about the strategy.
What should you compare during the forward test?
Keep a short checklist and review it weekly:
- Trade count versus what the backtest produced over similar periods.
- Win rate and average win/loss versus the backtest's trade log.
- Exit reasons. Are stops hitting far more often than expected? That can point to an intrabar assumption or a regime change.
- Exposure. Did the agent ever hold more concurrent positions than you planned for?
- Drawdown relative to the backtest's worst periods.
Big divergences are information. If the paper record diverges sharply from the backtest in ways the market regime does not explain, the backtest may have been flattering, and the Deflated Sharpe Ratio is a good place to recheck.
Paper trading TradFi and multi-strategy setups
Paper agents are also where you can run markets that are not yet available live. Gold, silver, oil, the S&P 500 and single stocks such as NVDA or TSLA trade on Hyperliquid through builder-deployed perps under HIP-3. On Trigr they can be backtested and run on paper agents, but live TradFi execution is not available in the current beta, so any agent with a TradFi slot must be a paper agent. The strategy side is covered in trading gold, the S&P 500 and NVDA on Hyperliquid.
Paper is also the right place to test a portfolio of strategies. An agent can hold up to 52 strategies, one per market. Running that combination on paper shows how the slots interact in real time before any money is committed. If allocations add up to more than 100%, you also see which entries get skipped when the free margin runs out. See multi-strategy crypto portfolio agents for how to choose the mix.
How many paper agents can you run?
Paper agents are included on every plan: 5 on Free, 25 on Trader and unlimited on Pro. That makes it reasonable to run several variants side by side, for example the same strategy at two different stop distances, and compare forward records rather than arguing from backtests alone.
If you work with an AI assistant through Trigr's MCP server, it can create a paused paper agent with up to 52 strategies, then rename or pause it. The performance seed cannot be changed after creation. It cannot start the agent; moving to real money is always your step.
When is a paper agent ready to go live?
There is no universal threshold, but a reasonable bar looks like this:
- The forward record has enough trades to compare with the backtest, and the comparison is within the range you would expect after costs.
- Nothing surprising happened with exposure, exits or data.
- Every entry would clear Trigr's $11 notional floor for Hyperliquid at the capital you plan to use, so no order gets raised above what your sizing rule intends.
- You have decided in advance what drawdown would make you pause the live agent.
Paper results, like backtests, are not guarantees, and live perpetual futures are leveraged and can lose more than expected. Going live at a fraction of your intended size is a sensible final forward test.
Next steps
Set up your first paper agent using the trading agents documentation, then follow from backtest to live on Hyperliquid when the forward record holds up.