In civil disputes, the plaintiff must prove his case by the preponderance of the evidence. To reach this standard, the plaintiff accumulates evidence by combining facts. I compare two models of this process. Decision makers can adapt their behavior for improved results, as assumed in some psychological models. Adaptive models predict that court practice will allow the plaintiff to combine facts according to relatively simple rules. Alternatively, decision makers can optimize their behavior for best results, as assumed in most economic models. Optimization models predict that court practice will require the plaintiff to combine facts in ways that conform to the laws of probability theory. The two predictions contradict each other when simple, adaptive rules violate the laws of probability theory. I show that actual practice in a California court allows the plaintiff to combine facts according to relatively simple rules that sometimes violate the laws of probability theory. Adaptation is, consequently, a better descriptive theory than optimization. Procedures that violate the laws of probability theory, however, are vulnerable to withering criticism. Given that trials proceed with deliberate speed under expert guidance, suboptimal adaptations are irrational. Optimization, consequently, is a better normative theory than adaptation.