In the old world of strategic intelligence, knowing how many tanks or drones a nation could build required a mole in the ministry or a satellite parked over a factory. Shadows and signals. A closed information game with very expensive chips.
A February 2026 study from Masaryk University changed the question. The fog of war has been pierced not by espionage but by something more embarrassing to the state: probabilistic hardware auditing.
The methodology treats the battlefield as a massive high-entropy dataset and reverse-engineers industrial capacity from it. Researchers stopped listening to what a state claims it can build. They started calculating what it actually builds by analyzing the scraps it leaves behind.
The Probabilistic Audit
The core technique is serial number analysis scaled to the extreme.
Not government press releases. Hardware loss distribution. Researchers scraped images of destroyed or captured equipment from social media and OSINT repositories like Oryx, then ran those through a share-of-loss model. When a specific version of a drone or vehicle stops appearing in battlefield loss data, that is hardware-level confirmation that the production line failed — regardless of what official communications say.
It is a massive SQL query against reality. The database is wreckage.
Fact-Checking the Firmware
The Masaryk study found a systematic discrepancy: official claims often suggest industrial output three to four times higher than what the OSINT data shows.
The state is just another vendor. Vendors lie about their specs. By using open-source information, researchers performed a real-time integrity check on national propaganda — treating the military-industrial complex like buggy firmware and using battlefield telemetry as the debugger.
The performance metrics that show up are not the ones in the press release.
