This project developed novel deep learning technologies to process historical Japanese economic data at scale: deep-learning powered layout analysis to recognize document layouts; a novel OCR architecture that can accurately transcribe over 13,000 Japanese characters with minimal training data; and novel methods for record linkage powered by both vision transformers and transformer large language models. These methods – which apply not just to Japanese but to a diversity of settings – have been publicly released as easy-to-use open-source packages, that are now being widely used by social scientists. We also published the methods in leading machine learning proceedings. We used these methods to construct large-scale economic networks (e.g., customer-supplier networks, board of directors, shareholders, geospatial, and familial) for the historical Japanese economy, using detailed firm level records and individual biographies. We moreover digitized firm balance sheets, product information, and addresses. In addition, we digitized detailed data about the economic policies of the Supreme Command for Allied Powers in Japan. Finally, we examined how the economic policies of the Supreme Command for Allied Powers during the post-World War II period reconfigured economic networks and influenced the Japanese economy. These data allow us to examine historical economic policies at an unprecedented level of detail. They show that as some economic relationships were disrupted by policies that aimed to reconfigure the Japanese economy, other relationships adjusted to preserve the status quo. The overall effect is a remarkable persistence in the configuration of economic activity, in the face of large-scale policies that attempted to majorly reconfigure the economy as part of efforts to prevent future militarization. This illustrates that detailed, comprehensive firm and individual level data are necessary to study the full, diverse effects of policies targeting a particular facet of the economy.