Compare two backtests and ask a better follow-up

Compare matched AAPL RSI rules, inspect held-out evidence and drawdown, then share only the result context you choose with AI Analyst and return to your report.

Change one rule, keep the remaining assumptions matched, and examine held-out results before drawing a conclusion. A higher historical return alone cannot identify a reliable strategy.

Change one rule at a time

This example asks whether moving an RSI crossing threshold from 30 to 35 changes the historical result. Keeping the remaining settings identical makes the comparison more informative.

Keep the question and the results connected

Review both proposals. Dates, sizing, exits, costs, and the holdout split need to match before a difference can reasonably be attributed to the entry threshold.

What the completed comparison showed

The November 9 runs covered January 3, 2023–December 30, 2025, rather than the requested December 31 end date. RSI 30 produced seven trades versus nine for RSI 35. Whole-window trade-return sums were 20.20% versus 13.34%; compounded one-position drawdowns were 9.23% versus 22.85%. Both arms used the same calculation methods. The final 30% holdout contained only two trades per arm: return sums of 13.15% versus 16.01%, with compounded one-position drawdowns of 9.23% versus 6.36%. That small holdout difference does not establish that RSI 35 is better overall. Buy and hold returned 121.63% over the whole observed window, which is a different comparison from held-out trades or a sized account. The engine version was unavailable and the sample-size warning remained material. The follow-up reiterated these limits without running another test. This case used three AI credits for the proposal, four backtest credits for both runs, and one AI credit for the follow-up; usage varies.

Ask about the evidence, not just the winner

A focused follow-up asks for interpretation of the completed tests without silently expanding the experiment.

Understand what a holdout can and cannot do

A holdout reserves part of the historical window for a separate check. Inspect its trade count as well as its performance. A segment with few trades offers limited evidence even when the headline is attractive.

After you inspect a holdout and use it to choose another rule, it has influenced your research. Repeatedly searching the same history can reward chance. These two runs are an educational comparison, not a recommendation to trade either rule.

Private context can make a follow-up more relevant, but it does not make an incomplete history complete. Check the preview, source dates and any excluded fields. Keep the result’s limitations beside the conclusion.