A machine-learning project a Canadian university student built in high school has become the basis for the most complete public archive of a quarter-century of USAID program evaluations, after the Trump administration took the original database offline last year.
USAID’s Development Experience Clearinghouse dates back to 1975 and has grown to include more than 168,000 documents on the agency’s foreign aid work. Its website went dark in early 2025 without warning as Elon Musk’s Department of Government Efficiency moved through the agency, and workers reportedly hauled its servers entirely away.
Musk wrote on X at the time that he had “spent the weekend feeding USAID into the wood chipper.”
Flashback: Court Freezes Musk’s USAID Shutdown on Constitutional Grounds
DOGE’s dismantling of USAID was part of a broader campaign that Musk once said could cut $2 trillion from the federal budget. That figure fell to $1 trillion, then $150 billion by the time Musk briefed Trump’s cabinet in April 2025, and by May the administration had formalized just $9.4 billion in cuts through Congress, including money pulled from foreign assistance.
Related: DOGE Spending Cuts Fall Far Short of Musk’s $2 Trillion Promise
NPR calculated that for every dollar the government had spent since the previous October, DOGE could claim credit for saving about four cents. Trump quietly shut down the unit in November 2025, months ahead of the July 2026 deadline set in his own executive order, without DOGE ever releasing a public accounting that outside experts could verify.
Ketevan Gallagher had no connection to that fight when she downloaded more than 4,000 evaluation PDFs from the DEC in 2024, for a senior research project using machine learning to analyze the reports. She’s now studying computer engineering at the University of Toronto. She reached out to AidData, a development research lab at William & Mary, when the DEC went offline the following year, to see if it could preserve her collection.
AidData published the archive this month at usaid-archive.aiddata.org, describing Gallagher’s organized dataset as “a unique solution for those searching for lessons learned” from decades of US foreign aid work. It now holds roughly 4,500 evaluations spanning 2000 to 2023, a subset of the original collection. Gallagher said she was “very grateful that AidData was able to preserve this data.”