The product of meta-orgs is taste. If boardgamegeek thinks that Twilight Struggle is a good game, then you, not having played it, should expect that it's likely a 'good' game. If Givewell thinks that AMF is a good charity, then you, not having looked at it yourself, should expect that it's likely a 'good' charity. With games that many people play, a website can average together those ratings and then sort them to generate a solid taste measure. With charities that have done things in the past and have models of what they can do in the future, an organization can evaluate the things done and the models for how things would change and estimate impacts and then sort by those impacts. But with scientific projects, this seems much more difficult, because you're extrapolating past the fringes of current knowledge. It's not "which board games that already exist are best?" but "which board game would be best, if you made it?" This is a skill that people have to various degree--someone is designing these things, after all--but I think it's very difficult to communicate, and more importantly, for listeners to have a good sense of why they should or should not trust another person's taste. Another way of looking at this is, with medical research, all of the cost-effectiveness is driven by whether or not the technology works. If the research is to validate or invalidate a theory, the usefulness of that theory (and thus the evidence) is determined by the technologies enabled by that theory (or the attention spared to work on other technologies that do work). But this is the thing that, by definition, we don't know yet, and is the thing that, say, SENS leadership spends its time thinking about. Do we approve this grant, or that grant? (This comment may oversell the degree to which tech growth is about creating new knowledge / tech rather than spreading old knowledge / tech, but I think the point still gets at what you're talking about here.)