Verify the strategy evidence.Then find the failure conditions.
Return is only the result. AlgoRover keeps provenance, data windows, cost assumptions, reproduced results, and stress tests in one record so you can see where a result came from and where it may fail.
How a strategy is reviewed
Every strategy follows the same sequence so the most attractive result is not the only result that survives.
- Output: a traceable strategy version
Record provenance and version
Keep the creator, research source, code or package version, and parameter boundaries.
- Output: comparable backtest assumptions
Establish the baseline
Record the data window, interval, starting capital, fees, slippage, and costs not yet modeled.
- Output: a repeatable result
Reproduce historical results
Run the same configuration on closed historical data and retain return, drawdown, and risk-adjusted metrics.
- Output: known limits and failure conditions
Apply deliberate stress
Compare in- and out-of-sample results, raise costs, add execution delay, and isolate adverse regimes.
What a stress test actually changes
Stress testing is not another run of the same backtest. It deliberately weakens the assumptions that helped the baseline.
| Dimension | Baseline record | Stress change | Question |
|---|---|---|---|
| Data | Sample window and market cycle | Out-of-sample data and different windows | Does the result reproduce across periods? |
| Costs | Fee, slippage, and funding assumptions | Higher friction and previously missing costs | Can the edge survive realistic friction? |
| Execution | Signal frequency and fill conditions | Delay, missed fills, and worse execution | Can it still operate under real constraints? |
| Regime | Trend, volatility, and liquidity | Adverse and extreme environments | Does it depend on one market state? |
Normalize the metric definitions first
The same name does not guarantee the same calculation. Strategies are comparable only when their definitions are visible.
- Return
- Net change from starting equity to ending equity.
- Sharpe
- Annualized mean periodic return divided by volatility; the current baseline uses a zero risk-free rate.
- Max Drawdown
- Largest peak-to-trough decline in the equity curve.
- Win Rate
- Profitable share of active holding intervals, not per-trade win rate.
How third-party strategies enter the library
Reconstructed public research retains its original source. Custom strategies submitted by KOLs retain their creator and version and are listed only after AlgoRover reviews their provenance, version, assumptions, and known limits.
Submit provenance and version
Record the creator, original research source, code or package version, and the supported parameter range.
Review, reproduce, and stress
Reproduce the historical result on a common baseline, identify cost or data gaps, and test out-of-sample behavior and failure conditions.
List with an explicit source status
Keep the third-party label and publish the verified scope, unknowns, and known limits without presenting the result as independent AlgoRover validation.
Stress tests expose fragility. They do not prove safety.
They cannot remove survivor bias, model risk, or future market change, and they cannot decide position size or whether a user should run a strategy. Their value is turning hidden dependencies into inspectable facts.