MilitarySpend
Independent defense economics research, published and maintained by machines.
MilitarySpend is an independent defense economics publication that presents military spending as live, readable pages rather than PDFs and spreadsheets. It runs to roughly 280 prerendered pages: 100 country profiles, all 50 US states and DC, defense contract breakdowns, scenario models and a reference desk for journalists. Agents handle the writing and upkeep, led by one that rewrites the Iran war pages every morning, and the live layers read public infrastructure directly rather than through a database.
Problem
Military spending is documented in public and read by almost nobody. SIPRI publishes the figures, but they arrive as PDFs and spreadsheets, and a billion dollars in a cell means nothing to anyone. The data needed a form people would actually stay with.
Solution
One agent does most of the work. At 9am every day it rewrites the Iran war pages — counters, the daily briefing, the timeline — reading the latest reporting and reconciling it against what the pages already say. It runs on an Opus-level model, because the judgment is the hard part: what actually changed, what is confirmed, and what is still only a claim.
The rest of the system runs behind it, less often:
- Publishing pipeline — seven stages: scout, writer, editor, SEO, quality gate, sync, link weaver - Freshness scanner - Headline curator - Maintenance editor — revises weak sections of older briefings - Glossary builder — monthly - Quarterly digest
The live layer reads public infrastructure directly:
- OpenSky — ADS-B aircraft positions - CelesTrak — satellite elements, propagated client-side - USAspending.gov — federal defense awards - GDELT — twenty-four-hour conflict event geolocation - SIPRI — the spending workbook everything else is projected from
There is no database. Every page is a static data module, prerendered through headless Chromium so the head tags flush before the snapshot is written.
Every run is recorded on a public status page. The remaining human job is judgment — checking that the machine is still writing things that are true.
Outcome
Around 30,000 views a month, rising to 2–3k on a day when something happens. It ranks first on Google for military spending, and Google's own AI cites it often enough to register in the traffic. Over 100 researchers and reporters subscribe, with a weekly digest assembled and sent automatically. The archive opened with 97k views on Reddit, on no advertising budget.
Features
- Real-Time Spending Counter
- Iran/US War Cost Tracker
- 100 Country Profiles
- All 50 US States + DC
- Live Aircraft Radar (OpenSky ADS-B)
- Military Satellite Tracker (CelesTrak + SGP4)
- Federal Contract Feed (USAspending.gov)
- Scenario Simulations
- Interactive World Map
- Conflict Cost Index
- Auto-Generated Research Blog
- Email Newsletter
- For Journalists Page
Lessons
- Timing beats polish every time.
- If you make boring data visual and real-time, people suddenly care about it.
Stack
- Frontend: React, D3.js, Three.js
- Backend: GitHub Actions, OpenRouter API
- Other: Grok 4.1 Fast, SIPRI Data, CSIS Data, USAspending.gov API, OpenSky Network, GDELT, CelesTrak / SGP4