Mindhacking idea: fudge the historical results such that the trend line always appears positive, and use that as a primary intervention. This requires either violating informed consent (a bad thing) or exercising doublethink or another method of self-deception.