Public records requests are a central mechanism for government transparency. In practice, they are slow, complex processes that require analyzing large amounts of messy, unstructured data. In this paper, we introduce RequestAtlas, a system that helps investigative journalists review large quantities of unstructured data that result from submitting many public records requests. RequestAtlas was developed through a year-long participatory design collaboration with the California Reporting Project (CRP), a journalistic collective researching police use of force and police misconduct in California. RequestAtlas helps journalists evaluate the results of public records requests for completeness and negotiate with agencies for additional information. RequestAtlas has had significant real-world impact. It has been deployed for more than a year to identify missing data in response to public records requests and to facilitate negotiation with public records request officers. Through the process of designing and observing the use of RequestAtlas, we explore the technical challenges associated with the public records request process and the design needs of investigative journalists more generally. We argue that public records requests represent an instance of an adversarial technical relationship in which two entities engage in a prolonged, iterative, often adversarial exchange of information. Technologists can support information-gathering efforts within these adversarial technical relationships by building flexible local solutions that help both entities account for the state of the ongoing information exchange. Additionally, we offer insights on ways to design applications that can assist investigative journalists in the inevitably significant data cleaning phase of processing large documents while supporting journalistic norms of verification and human review. Finally, we reflect on the ways that this participatory design process, despite its success, lays bare some of the limitations inherent in the public records request process and in the ''request and respond'' model of transparency more generally.

String matching is at the core of data cleaning, record matching, and information retrieval. String matching relies on a similarity measure that evaluates the similarity of two strings, regarding the two as a match if their similarity is larger than a user-defined threshold. In our collaboration with journalists and public defenders, we found that real-world datasets, such as police rosters that journalists and public defenders work with, often contain acronyms, abbreviations, and typos, thanks to errors during manual entry, into, say, a spreadsheet or a form. Unfortunately, traditional similarity measures lead to low accuracy since they do not consider all three aspects together. Some recent work proposes leveraging synonym rules to improve matching, but either requires these rules to be provided upfront, or generated prior to matching, which leads to low accuracy in our setting and similar ones. To address these limitations, we propose Smash, a simple yet effective measure to assess the similarity of two strings with acronyms, abbreviations, and typos, all without relying on synonym rules. We design a dynamic programming algorithm to efficiently compute this measure, along with two optimizations that improve accuracy. We show that compared to the best baselines, including one based on ChatGPT with GPT-4, Smash improves the max and mean F-score by 23.5% and 110.8%, respectively. We implement Smash in OpenRefine, a graphical data cleaning tool, to facilitate its use by journalists, public defenders, and other non-programmers for data cleaning.