Run a reproducible strategy review
Goal
Move from an idea to a documented review without changing the running bot.
Lesson
Use the experiment record from the knowledge library. Keep candidate code and configs isolated. Run the same fixed dataset and settings for the baseline and candidate. Test signal behavior and lookahead before interpreting returns. Save complete output files and source hashes.
When a result changes, identify whether the cause was code, data, settings, dependencies, costs, or market period. Keep failed experiments in the record. A negative result narrows the next question; it does not justify hiding the run or repeatedly searching the same sample.
Practice
Prepare a one-page review containing the hypothesis, exact command, setup, result, comparison, validation status, and limitations. Ask a second reviewer to reproduce it.