Compare one change

Prerequisites

Have a completed Agent Lab baseline using the first SJS strategy. Retain the exact source revision, data identity, and execution settings. The same method works with saved app historical results if you record those inputs explicitly.

Choose a single hypothesis

For this exercise, change only the crossover quantity from one to two. Keep instrument, interval, bounds, timezone, session options, fill delay, and all other source fixed. Use sufficient simulated capital for both cases and no additional quantity gate.

The hypothesis is that signal times remain identical because quantity does not participate in the crossover calculation. Executed outcomes can still differ through capital or other execution checks.

Save the candidate and run

Save a new revision, run the same historical window, and retain both run identities. In Agent Lab, choose the baseline with Use as baseline and compare it with the candidate.

Do not edit the window to make the candidate look better during this comparison. If you need another window, treat that as another experiment with its own baseline.

Inspect the first difference

If signals differ, inspect source and input differences before interpreting totals. If a larger request is rejected, the hypothesis about execution size did not establish an equivalent fill path.

Record the conclusion

Write what changed, what stayed fixed, the first changed event, and any limitation of the simulator. A higher total alone is not an explanation or a prediction.

Common mistakes are changing quantity and fill delay together, comparing different data windows, or publishing the candidate without checking its selected revision.

Next: add L3 oversight.