Anyone can check the first fact in this story in about fifteen seconds. USAspending.gov’s spending profile for federal account 097-0850 — the Payment to the Department of Defense Medicare-Eligible Retiree Health Care Fund — reports $37,108,000,000 obligated for fiscal 2026 against total budgetary resources of $18,554,000,000. That is exactly two hundred per cent, to six decimal places. An account cannot obligate more than it has; the figure is impossible on its face.
The larger of the pair does the same thing less tidily. Account 097-0040, Payments to Military Retirement Fund, shows $322,000,000,000 obligated against $161,413,000,000 available — 199.49 per cent, an $826m miss from a clean doubling. That account did hit exactly two times in the periods running from February to April, when its obligations figure read $322,826,000,000; the resources side has since been updated and the obligations side has not.
Both accounts behaved normally until this year. In fiscal 2023 and 2024 each sat at 100.00 per cent of its resources, and in 2025 at 99.75 per cent — which is what these accounts should look like, since they are mandatory transfers that obligate their full appropriation.
Behind the account pages is a reconciliation the Treasury’s Bureau of the Fiscal Service publishes for every agency. Departments submit two files: File A, drawn from their own Standard Form 133 budget execution reports, and File B, the object-class and program-activity detail. The two are supposed to agree. For the Defense Department in period 9 of fiscal 2026 they differ by $373,136,506,601.13 — a figure USAspending reports itself, in a panel built for the purpose.
The shape of that gap is the part nobody has published. Of the 321 Defense accounts with a difference, 44 are an exact integer multiple to the cent — 17 where File B is twice File A, 10 at three times, 12 at four times and five at five times. Widening the test slightly, 35 accounts at or within two cents of an integer multiple carry $348,125,481,656.26 between them, which is 93.3 per cent of the department’s entire gap. Whatever produced this did not produce noise; it produced multiplication.
The multipliers are also stable. Account 097-0850 has been at exactly two times, 097-5472 at exactly three and 097-1612 at exactly four in every one of the seven monthly submissions from period 3 to period 9. A transient upload error does not hold a constant factor for seven months.
A control year makes the change legible. In period 9 of fiscal 2025, the same department filed 382 accounts with not a single integer multiple among them, a worst-case discrepancy of $47,449,438.00, and a net position of $4.37bn in the other direction. The reconciliation series itself runs back to period 6 of fiscal 2017, 82 agency-period records, so this is not the trailing edge of a short file.
It is also close to a Defense-only phenomenon. Of 111 agencies reporting for period 9, 24 have a non-zero difference; the 23 that are not the Pentagon net to a combined $1.92bn — less than one per cent of the Defense figure, and in surplus rather than deficit.
Why it is happening is not something these files answer. Agencies file assurance statements alongside their submissions explaining known exceptions, and USAspending’s API publishes a link to the Defense Department’s for period 9 — but that link returns a 404, as do the recent assurance statements for every agency checked, so the explanation may exist without being reachable. The most recent Defense statement that still resolves is from fiscal 2022, and it commits the department to File A matching its SF-133 data. The department has not said anything about the current gap, and USAspending’s own methodology page notes only that differences of this kind are expected to be explained by the agency.