Bryan Minor, Privateer; Matthew Lugo, Privateer
https://doi.org/10.64861/BPYH6248
Keywords: Space Domain Awareness, Open-Source Intelligence, Event Forecasting, Large Language Models, Retrieval-Augmented Generation, Predictive Analytics, Maneuver Detection, Data Fusion, Artificial Intelligence, Threat Assessment, Probabilistic Calibration
Abstract:
Space Domain Awareness must prioritize limited sensing and analyst resources across a rapidly expanding satellite population. This study evaluates whether PAI (Publicly Available Information) can be collected at scale through MCP (Model Context Protocol) servers and web search, then transformed with LLM agents into Standard or Enhanced monitoring recommendations. The pipeline resolves satellite identity, normalizes heterogeneous evidence, applies strict historical cutoffs, extracts attributable behavioral cues, and preserves provenance from retrieval through classification. A conservative policy elevates satellites only when verified prior concerning behavior, strong object-specific operational cues, documented orbital changes, or independent evidence supports greater attention. In an early 32-case experiment, the system processed 1,737 normalized records, but admitted only 37 decision cues after identity, timing, content, duplication, and source-independence controls. It recommended Enhanced monitoring for 10 cases and Standard monitoring for 22. Among 29 cases with hidden gold actions, all 5 controls remained Standard, while 8 of 24 historical positives were elevated, demonstrating high precision but limited capture. The results show that scalable PAI analysis is feasible, but identity resolution, historical source coverage, and persistent provenance are fundamental. This approach can help operational SDA focus collection on satellites whose behavior and context warrant additional scrutiny.
Date of Conference: September 15-18, 2026
Track: Space Domain Awareness