How does employer learning affect the allocation of talent in the market for research scientists? I study this question using the job histories of 40,000 Ph.D.‘s in computer science (CS) matched to their scientific publications and patent applications. Authorship of a CS conference proceeding doubles the probability that a researcher moves to one of the top tech firms in the following year, controlling for her origin firm and experience, implying a strong role for public learning in the matching process between more productive workers and more productive firms. Many higher-quality papers are accompanied by a related patent application, but the existence of an application is private information for 18 months. Authors of such papers are somewhat less likely to move up the firm ladder in the following year, but are more likely to end up at a top firm within three years, as predicted by a model of employer wage setting with asymmetric information. I estimate a structural version of the model and find that in the absence of employer learning from scientific publications, the innovation output of early-career computer scientists would drop by 16%. Disclosing patent applications one year faster would increase innovation by 1%, driven by a faster rate of assortative matching.