Build the graph
Papers become directed claim edges between coded economics concepts. This is the only stage where AI is allowed.
How FrontierGraph works
The system is built around one hard constraint: AI may help convert text into graph structure, but the ranking itself should remain deterministic, inspectable, and open to scrutiny.
Score missing links
The engine combines gap signals, path support, motif completion, hub penalties, field distance, and prior contact.
Inspect the evidence
Users move from a shortlist to mediators, paths, supporting papers, and exportable idea briefs.
Signals
Underexplored concept-pair gaps, path-implied missing direct edges, motif completion, low prior contact, field distance, and hub penalties.
Current scope
No user accounts, no saved agendas, no downstream impact forecasting, and no custom frontend beyond Streamlit yet.
Tool, not oracle
FrontierGraph is designed to assist judgment, not replace it. It surfaces structured opportunities and exposes the evidence.