Compare strategies with Multiverse Agent
Agents → Compare runs selected native spot strategies in separate virtual wallets. Each receives the same closed candles, starting capital and cost assumptions, so you can inspect how different strategies behaved over a shared period.
The comparison engine does not require an AI provider. Explain with AI uses the provider connected in Agent settings. A comparison does not place live orders or automatically change the strategy assigned to your pair.
Set up a comparison
- Select the Gunbot instance, then open Agents → Compare.
- Select a configured spot pair.
- Choose a different strategy in each dropdown. Use Add to add more selections; leave unused selections empty.
- Set the shared candle period and inspect the simulation parameters. Review the starting capital, order sizing, fee and slippage assumptions.
- Start the comparison using Compare selected strategies.
Strategies use their own settings. The active preset includes its pair overrides; alternatives do not automatically inherit those overrides. Parameter edits inside the simulation do not change live presets. Edited rows are identified as simulation edited.
The dropdown catalog is broader than verified simulation coverage. A strategy requiring missing historical inputs or an unavailable adapter reports an error. Other completed simulations remain visible, but a failed selection cannot support a complete comparison. Futures, custom JavaScript, order-book and external-signal workflows require their own execution models or historical inputs; do not assume a spot candle comparison covers them.
Read the results
Start with the shared chart and strategy ranking, then inspect individual charts and numeric results:
- Return and equity: the simulated result, including open virtual holdings valued at the observed market price.
- Drawdown: the decline from a previous equity peak. A higher final return can come with a deeper drawdown.
- Execution counts: how often each candidate bought or sold. These counts are not automatically matched round trips.
- Sizing and costs: check that the strategy traded a meaningful portion of the common virtual capital. Tiny sizing can produce a very small return even when orders were executed.
The direct comparison identifies the observed return leader and lower-drawdown strategy for each completed pair of selections. If both lost value, the return leader may simply have lost less. A historical leader is not a prediction of the best future strategy.
Fee and slippage inputs are percentages per fill. Slippage is modeled as an additional transaction-cost allowance, rather than a simulated adverse execution price or market impact.
Separate replay from forward observation
Historical replay covers the retained closed candles available when the observation starts. Forward observation adds new complete candles after that boundary, continuing each virtual strategy’s holdings.
Read the two periods separately. A strategy that leads the historical replay may not lead the forward period. The displayed forward score is return minus maximum drawdown in percentage points; it is a comparison rule, not statistical confidence.
Stale candles, changed settings, missing inputs or an outdated simulation invalidate results. If the interface asks for a fresh observation, restart after checking the reason. Restarting resets the forward boundary; stopping retains results. Background observations resume only in the supported observation lifecycle, so reopen the comparison after a Gunbot restart.
Ask AI to explain the current result
With a provider connected, choose Explain with AI and ask a specific question:
Compare return, drawdown and trading activity. Which strategy led each observed period, and what makes this evidence incomplete?
The agent can discuss the actual comparison evidence. It cannot establish future profitability or turn a small sample into a reliable winner. Read the charts, sizing, cost assumptions and missing-data messages alongside the explanation.
Continue with a reviewed decision
Multiverse Agent does not automatically switch live strategies. Use the result to decide whether a configuration deserves further testing or a manual change. If you excluded or edited the active baseline, treat the comparison as simulation-only evidence rather than a verified live-switch recommendation.
Use Backtesting for the broader testing toolkit, or Simulator mode to observe a spot setup against current market data with simulated execution.