What the public record shows about FEMA Public Assistance money
ObliGate runs a small research program on the public FEMA Public Assistance record: public datasets, public code, observational measurement, and no causal claims.
Every figure below carries the data vintage it was computed from, and when a source dataset changes underneath a published number, we correct it in the open and version the correction rather than editing quietly.
FEMA pulls back about 1 in 10 Public Assistance funding actions ($44.3 billion), and it concentrates in complex projects and nonprofits.
Using FEMA's own public data (1,220,585 project records, August 2026 pull), obligated recovery dollars get revised downward far more than the program's headline numbers suggest: 10.0% of funding actions are deobligations, worth $44.3B gross, or 13.5% of the $327.9B in gross obligations.
Water-control, building, and utility projects, the documentation- and procurement-heavy ones, are clawed back up to about 2.4x as often as roads and emergency work, and nonprofits (11.8%) fare worse than governments (9.8%). And it is slow: about 1 in 8 projects is deobligated at some point, a median of nearly three years after the money is first obligated.
The reason for each pullback is not recorded in the data, so we do not call it all "clawback." The pattern is exactly the post-obligation reconciliation risk ObliGate exists to reduce.
When applicants fight back and win, the money is visible in the record.
In FEMA's legacy-era record (pre-2016 disasters), granted appeals were followed by new positive obligations on 85% of projects versus 14% of denied controls, consistent with decisions being implemented. Where money was measurable against the pre-appeal clawback, outcomes varied widely: median 63% of contested dollars restored, a fifth of winners near zero, a quarter roughly in full.
The denied arm is the control. If new obligations after a decision were just ordinary amendments, denied projects would show them at a similar rate; instead they sit at a 13.6% floor, and their movement arrives later and larger, which is the amendment signature rather than implementation.
This is decision implementation made observable, not evidence that appealing pays. Applicants select which denials to contest, and FEMA decides which appeals to grant, so neither selection is modeled here. The findings describe obligation events, which are funds made available, not cash drawn down.
The methods note documents the join protocol between two public FEMA datasets, the falsification-controlled restoration measure, and the era-coverage map that shows where the public financial record supports this kind of accountability.
Fisher, A. (2026). "Linking Clawbacks to Appeals in FEMA Public Assistance: A Join Protocol and a Falsification-Controlled Measure of Post-Appeal Restoration." OSF.
Both figures, the method, the churn versus reduction bound, and the limitations, on one page.
Read the one-pagerThe falsification control, the per-arm timing and dollars, the era-coverage map, and the limitations.
Read the one-pagerThis research is produced by ObliGate Compliance Inc., which builds documentation software for FEMA Public Assistance applicants. It is observational analysis of public records, and it offers no advice to any applicant on whether to appeal.
No causal claim is made anywhere in this work. Selection into contesting a determination, and into winning one, is unmodeled by design. Obligations are funds made available, not cash drawn down. OpenFEMA is raw administrative data that FEMA restates over time, which is why every figure here carries its pull date and row count.