The United States Agency for International Development (USAID) shut its doors a year ago. Among the assets at risk: a quarter-century of evaluation reports documenting what worked, what didn't, and why. We've archived the entire corpus of knowledge, which consists of ~4,500 English-language performance and impact evaluations published between 2000 and 2023. They're free, ungated, and searchable at USAID-archive.aiddata.org.
This blog is the first in a two-part series on endangered data resources. With this piece, we detail AidData’s collaborations to save a key archive at risk of disappearance. In Part 2, we will explore findings from AidData’s recently-published Data Asset Inventory, which examines resources that are at risk of meeting the same fate as the USAID Development Experience Repository.
What’s in the USAID archive
The archive contains a curated set of USAID evaluation reports, capturing decades of evidence on foreign aid programs. They include performance evaluations that measure the extent to which USAID programs generated their expected outputs, as well as impact evaluations that provide evidence of the net, attributable impacts of USAID programs between 2000 and 2023.
USAID was a leader in evidence-based decision-making, accountability, and organizational learning. For decades, it commissioned program evaluations and maintained a documentation repository, but its internal evaluation culture tended to emphasize compliance over learning. Then, in 2011, the agency adopted a new Evaluation Policy designed to improve both the quantity and quality of evaluations through a more demanding and explicit set of requirements. The policy required that evaluations be built into program and project design from the outset, rather than added at the end of implementation, and it emphasized answering strategic questions about program impacts. It also required USAID operating units to systematically plan evaluations and ensure they were adequately resourced and methodologically rigorous. While it did not mandate a fixed share of program budgets for evaluation, it effectively normalized evaluation as a standard cost of development programming and reinforced a broader shift toward an organizational culture of learning. The policy also strengthened transparency expectations by requiring that evaluation findings be documented, disseminated, and made publicly available, enabling both internal and external stakeholders to learn from results.
Although open data is a public good, USAID’s Development Experience Clearinghouse (the DEC) was not very user-friendly. It was called a pdf graveyard, the kind of data graveyard that AidData said you should avoid. But it was openly accessible, which meant that someone somewhere was probably using it as training data for an AI algorithm.
How it came to be
At least one of the people using USAID’s Development Experience Clearinghouse to run an algorithm: a high-school student named Ketevan Gallagher, who is now studying at University of Toronto. As a part of her senior research project, Ketevan collected over 4,000 pdfs of evaluations from the DEC for use in a machine learning project.
After the DEC went offline, Ketevan reached out to AidData to see if we could archive her work. When we sifted through the data that Ketevan had cleaned and structured, we saw an opportunity to provide a unique solution for those searching for lessons learned from evaluations of U.S. foreign assistance programs.
There are other efforts underway to derive insights from the broader corpus of DEC documentation—including a smorgasbord of training manuals, handbooks, literature reviews, workshop and conference summaries, strategy documents, policy papers, feasibility studies, engineering designs, social and environmental assessments, political economy analyses, contractor deliverables, audits, and compliance reviews—using AI or Large Language Models (LLMs). These include DevelopMetrics’s use of a dedicated LLM to preserve insights from the DEC and the archival work of the Humanitarian Archive Emergency project at the University of Manchester. However, none of these efforts are primarily focused on retaining evaluation reports for open research access. USAID-archive.aiddata.org is free, ungated, and exclusively focused on providing evaluation reports.
To make the search experience more user-friendly, AidData has made all of the underlying PDF documentation publicly accessible. They represent the corpus of knowledge that we are preserving.
Why this evidence matters
The archive matters beyond specialist circles. Most Americans vastly overestimate the scale of U.S. foreign aid—and that misperception shapes whether this kind of evidence is seen as worthy of preservation.
Nearly nine in ten adults (86%) in a 2025 KFF Poll overestimated the share of the federal budget spent on foreign aid, with respondents saying, on average, that the U.S. spends about a quarter (26%) of its budget on foreign aid. In fact, less than 1% of the federal budget is earmarked for economic-related foreign assistance.
The perception gap is understandable. Aid sounds big because it is described in sweeping, global terms—for example, 92 million lives were saved through U.S. health assistance over the past two decades.
Yet, in the context of other global problems that public spending is expected to address, aid is quite small. The U.S. foreign assistance budget ($40-60 billion a year) is orders of magnitude smaller than the Department of Defense’s budget ($961 billion in fiscal year 2026).
U.S. foreign assistance may be renamed, reduced, reprioritized or reorganized, but Congress has made clear that it is not going away. Moving forward, access to reliable data and evidence will become even more important to legislators who want to convince skeptical voters that foreign aid is effective.
A hard truth for both USAID cheerleaders and faultfinders is that sometimes aid effectively accomplished its intended purposes, and sometimes it didn’t. The question is: how do we know what worked, what didn't, and why?
Lessons learned need to be publicly accessible. USAID-archive.aiddata.org is currently the only portal of DEC evaluation reports that does not require log-in credentials. We have sought to make it as easily findable, intuitive to use, and as machine-readable as possible.
USAID-archive.aiddata.org is one piece of a larger effort. We recently published an inventory of the most widely used development data assets across seven key sectors—a first step toward identifying what development data the world can afford to lose, and what it cannot. And through our Reimagining Development Data (RD²) project, funded by a $1 million grant from the Hewlett Foundation, AidData is working with those who fund, produce, and use development data to ensure that lessons learned get built upon—instead of disappearing.
As development finance evolves, the evidence base that informs it must evolve too. That starts with making sure we don’t lose what we already know.
