Is It Better to Have a Dedicated Process for Testing New Trading Ideas?

Trying a new concept directly within normal trading can make it difficult to determine whether it helped or hurt performance. How would you separate experimentation from established execution? What records would you keep so that a promising idea can be evaluated without disrupting your existing approach?