How Do You Decide When a Trading Experiment Is Finished?

Research can continue indefinitely if there is always another parameter to test. What criteria help you decide that an experiment has answered its original question and is ready to be documented?