OM.

These results cut deep. When students ask me how to choose a research project, I tell them to look at the intersection of three properties: rigor, relevance, and surprise. Good research is relevant and rigorous, but also unexpected. Otherwise, it just confirms something people already intuit. But surprise is not just a matter of taste. It is a core idea in information theory. An unexpected yet verifiable result carries more information than one we anticipated. As scholars, we are trained to argue that our results are at least a bit unexpected in the introduction of our papers. In social science papers, we usually do this using the classical trick of the two-handed economist. But an LLM forecast is very different from a narrative argument. It provides a quantitative expectation of what a study might find. Eva's results show that forecasts on treatment effects of pre-registered studies can be quite accurate. This suggests these forecasts could become part of the preregistration. In other words, we could start preregistering surprise. Of course, not all valuable research needs to be counterintuitive or surprising (this applies mostly to breakthroughs). Replication and precision matter too. But these forecasts could help us distinguish between results that confirm our expectations and those that challenge them. Great job on this, Eva! I think there's a lot of potential here.