There is a "replicability crisis" in the scientific literature. Scientists attempting to redo experiments in reputable, peer-reviewed journals have found that staggering numbers of these experiments up to 90% do not work. Patents, like scientific articles, contain experiments. These experiments often form the backbone of the patent and provide crucial support for patentability. Patent examiners use these experiments to evaluate whether the invention works, and thus whether the patent should be granted. The replicability crisis in the scientific literature is therefore of utmost importance to the patent system. Transferring the insights of the replicability crisis to patents begs the question of whether experiments in patents are similarly irreplicable-a question that has not previously been asked in the literature. This Article's novel empirical study of 500 patents and applications measures the replicability of experiments in patents. Using methodological quality of the experiment as a proxy for replicability, this study finds that experiments in patents have very poor methodological quality, which means that they are likely irreplicable at rates at least as high as experiments in scientific journals. Given the centrality of experiments to patents, this Article's empirical finding is a crisis not only of replicability, but also of patent law, and has important implications for patent doctrine, theory, and policy. Patent law relies on the assumption that, when a patent is filed, it has been "reduced to practice"-meaning that the invention works. The reality is that most inventions likely do not work, casting serious doubt on this assumption. Similarly, the underlying justification for patents-incentivizing innovation-requires that they contain enough information to teach others how to make and use the invention. Irreplicable patents may not do this. In short, there is a fundamental mismatch between patent law's theory and doctrine, which treat patents as reflecting fully formulated inventions, and what patents often represent, which is early stage, frequently unproven, results. The replicability literature-and the findings of this Article-teach us that experiments reported in patents are not reliable enough to merit the level of control and influence that they are granted in the patent system. This Article argues that patent law cannot solve irreplicability; instead, patent theory must better reflect the reality of irreplicability. Several policy proposals flow from this reformulation, including easing the process of invalidating inoperable patents, improving disclosure of ex post data in patents, and clarifying the experimental use exception.
