Prompt Details
Model
(claude-5-5-sonnet)
Token size
501
Example input
[strategy]: Trend following: buy when the 20 EMA crosses above the 50 EMA and sell when it crosses below, fixed stop of 1R, take profit at 1.8R
[market]: EURUSD, 1-hour chart
[results]: Period: January to December 2023. 42 trades, win rate 45% (19 wins, 23 losses), average win +1.8R, average loss -1.0R, net result +11.2R, maximum drawdown 6R
[parameters]: Only EMA 20 and 50 were tested; no optimization; no out-of-sample split
[costs]: Spread 1 pip, no commission, no slippage modeled; entries at the close of the signal bar
Example output
## Verdict
Limited evidence. The strategy is simple with few parameters, but 42 trades over one year is a small sample and slippage is not modeled.
## Consistency Check
- Win rate: 19 wins / 42 trades = 45.2%, which matches the stated 45%.
- Net result: 19 x 1.8R = +34.2R; 23 losses x 1.0R = -23.0R; total +11.2R, which matches.
- Expectancy per trade: 11.2R / 42 = +0.27R.
- Break-even win rate with a 1.8R average win and a 1.0R average loss: 1 / (1 + 1.8) = 35.7%. The observed 45.2% is above it.
- Sensitivity: removing three average-sized wins (3 x 1.8R = 5.4R) leaves +5.8R over 39 trades, or +0.15R per trade. Individual trades were not provided, so the average win is used as a proxy.
## Bias Checklist
- Look-ahead bias: Medium. Entering at the close of the same bar that produced the signal assumes you could trade that exact price.
- Overfitting: Low. Only two EMA lengths were used and nothing was optimized.
- Sample size: High. With 42 trades, a few trades can change the result materially.
- Survivorship or selection bias: Unknown. Only EURUSD is mentioned, and it is not stated whether other pairs were tested and dropped.
- Unrealistic costs or fills: Medium. A 1-pip spread with no slippage or commission is optimistic.
- Regime dependence: High. One calendar year covers a limited range of market conditions.
- Data snooping: Low. No variations were tested.
## Missing Information
- Whether other pairs or timeframes were tested before this one.
- Trade-by-trade results, to check whether a few trades drive the profit.
- Swap or rollover costs for trades held overnight.
## Next Tests
1. Re-run with entries at the next bar's open. Confidence rises if the expectancy stays clearly positive.
2. Run the same rules unchanged on other years. Confidence rises if results stay positive in different conditions.
3. Add slippage and a wider spread. Confidence falls if the expectancy approaches zero.
4. Resample the order of trades (Monte Carlo) to see the range of drawdowns beyond the observed 6R.
5. Test neighboring parameters, such as EMA 15/45 and 25/55. Confidence rises if results are similar.
## Questions to Answer Before Trusting the Results
1. Could you really have entered at the close price of the signal bar?
2. Were other pairs tested before this one was kept?
3. How many of the 19 wins came from a small number of large moves?
By purchasing this prompt, you agree to our terms of service
CLAUDE-5-5-SONNET
Audit your trading backtest before you trust it. Paste your strategy, results, parameters tested and cost assumptions to get a reliability verdict, recomputed metrics with the math shown, a bias checklist (look-ahead, overfitting, sample size, data snooping), what is missing, and a prioritized list of tests to run next. Built for forex, futures, stock and crypto traders. It reviews your process only: no signals, no profit predictions, no financial advice.
...more
Added 59 minutes ago
