STDN-GEN: rapid synthesis of layered critical-material dependency networks for supply-chain sustainability analysis

Assessing a critical-material supply chain first requires a map: which components make up a technology, which materials those components need, and which countries produce those materials. Building such maps is slow, manual work, which limits how many technologies an analyst can screen. We introduce STDN-GEN, a system in which several large-language-model agents draft these maps and reconcile their proposals against a curated vocabulary of component and material names. The output is an auditable four-level network from technology to component to material to producing country, built in minutes rather than days. Across 180 technologies in microelectronics, biotechnology, and pharmaceuticals, enforcing the shared vocabulary is by far the largest source of run-to-run reproducibility, raising agreement between repeated analyses of the same technology roughly fivefold; debate between agents helps in some domains and widens the set of candidate dependencies in others. The system recovers 98% of the components human annotators identified, and its maps reproduce the extreme production concentration that published assessments report for neodymium.