Traceable answers
in the mission stack
At SHINE 2026, the NATO ACT Innovation Continuum on-site experiment in Halifax, Canada, Evidence Graph ran on wargame material and served analysts and other agents through its MCP interface, with evidence attached to every answer.
The scenario: Arctic Fracture
Arctic Fracture is a fictional NATO wargame scenario set in 2032, written to support wargames and C4ISR technology demonstrations. A near-peer adversary uses weather, cyber attacks and proxy forces to seize ground in the Arctic, cut undersea cables and disrupt satellite links, then opens a second front to delay a NATO force deploying from Canada.
The constant pressure is C4ISR degradation in every phase: untrusted networks, degraded GPS, disrupted satcom, electronic warfare and contested intelligence fusion.
Covert maritime crisis
Proxy seizure, cable cuts, mines and submarines. Tests unclear intelligence and untrusted sensors.
Command collapse
A strategic cyber attack breaks command networks. Tests rebuilding C2 and data trust.
Article 5 response
Two-front operations, air defence and sea clearance. Tests multi-source fusion.
Arctic Fracture is a fictional exercise scenario; all countries, forces and events in it are fictional. The test reports are synthetic.
A corpus built to trip up an AI
51 synthetic reports, one per BLUE Arctic order-of-battle entry, covering the opening 30 hours of the Article 5 response. They are reports about reporting: deliberate information problems rather than combat narrative. A provisional surface contact, two differing summaries of one acoustic contact, an unclear warning notice and an unnumbered task group code.
Evidence view
Which organization does AF32-FE-BLUE-001 support?
North Support Cell [C-017]
C-017, report R-004 v2
CL-021: force element SUPPORTS organization
Integration
Users included ISR analysts, collection managers, C2 users, targeteers and AI agents in the mission stack, reaching Evidence Graph through a web interface, REST API or MCP server. Each received a checked answer with cited passages, graph claims and a release ID.
Proposed next step: answers drawing on coalition shared data (STANAG 4559 CSD) through a separate MCP service, with the same evidence chain. Retrieval support only; no automatic sensor tasking.