Methodology
How this site is made: one person, an AI research pipeline, and a paper trail.
MilitarySpend is a solo-run publication. The daily research is AI-assisted and every published change is reviewed before it ships. This page is the entire methodology — sources, ranges, updates, and corrections — with no imaginary newsroom attached.
Where the numbers come from
SIPRI anchors the global spending comparisons. For everything narrower — war costs, casualties, contracts — primary documents come first: budgets, appropriations texts, procurement notices, CRS reports, and official releases. Corporate filings back the contractor pages. Think-tank reports and wire coverage provide context, but they never outrank the direct public record.
Ranges, not false precision
Primary sources disagree, especially in wartime. Instead of blending conflicting figures into one confident-looking number, casualty and cost rows are published as low–high ranges with a confidence flag, a named source, and the date the figure was last verified. A claim backed by a single source stays labeled provisional until it is independently corroborated.
The daily update pipeline
Every morning an automated research assistant sweeps for developments affecting the live trackers. It is not allowed to change a figure without corroboration from multiple independent sources, and its output lands as a pull request that I review before anything publishes. Rows that were checked but unchanged get dated "verified, no change" annotations, so staleness is visible instead of hidden.
Reading the live counters
The counters interpolate between verified data points to make scale legible — they are estimates, not ledgers. Public data lags the events it describes, some defense spending is off-budget or partially visible, and rapid-turn figures are directional until the accounting record firms up. The pages say so rather than pretending otherwise.
Normalization
When figures are compared, year, currency, and scope are normalized first. Nominal and inflation-adjusted values are not mixed unless the text makes the distinction explicit.
Corrections and revisions
Factual errors are fixed as soon as they are confirmed. When a change would alter a reasonable reader’s takeaway, it is disclosed on the page rather than silently patched. Source challenges and correction requests go through the contact page — review is guaranteed, agreement is not.
Current core tracker assumptions
SIPRI baseline year: 2025
Tracker display year: 2026
Counter scope: the top six military spenders in the live homepage tracker, plus a broader public database for comparisons and context.
Every dataset behind the trackers is free to download on the Data page (CC-BY 4.0). For a plain-language walkthrough of the automation, see How It Works; the pipeline's public activity log lives at Site Status.