SwanForge

Inspiration Most disaster systems prepare for the failures we already expect. BLACK SWAN FORGE was inspired by the opposite question: what if the next citywide collapse starts from a small, boring dependency no one is watching? Modern cities are not separate systems. Power, water, telecom, hospitals, payments, transport, fuel, and emergency response all depend on each other. A weak link in one layer can silently disable another. We built SwanForge to expose those hidden chains before they become real disasters.

What it does BLACK SWAN FORGE is an adversarial resilience simulator for cities. It models a city as an interdependent network: [G = (V, E)]where each node is a critical service and each edge is a dependency. The system searches for small shocks that create unexpectedly large cascades, then recommends low-cost interventions that prevent collapse. The prototype can: simulate cascading failures across 28 city systems and 46 dependencies run adversarial search to find worst-case failure paths explain the causal chain behind a collapse compare interventions against random and heuristic baselines test robustness across uncertainty connect abstract dependency failures to physical grid behavior using an IEEE-14 power-flow model

How we built it We built SwanForge as a Next.js and TypeScript prototype with a separate simulation engine. The core engine includes: deterministic cascade propagation delayed dependencies and backup systems evolutionary adversarial search causal minimization intervention optimization multi-seed benchmark experiments uncertainty and sensitivity analysis IEEE-14 DC power-flow cascade modeling The UI turns the research into a live demonstration: users can run a 90-second demo, watch a quiet failure become a citywide cascade, and see which intervention prevents it.

Challenges we ran into The hardest part was making the system both ambitious and defensible. A “moonshot” idea can easily become vague, so we focused on making every claim executable. If we said the system finds hidden collapse paths, we made it run the search. If we said it recommends interventions, we compared those interventions to baselines. If we said the model connects to real infrastructure, we added a public IEEE-14 grid test. Another challenge was balancing clarity with depth. The project needed to be understandable to judges while still showing real technical substance underneath. Accomplishments that we're proud of We are proud that SwanForge is not just a dashboard or pitch concept. It is a working prototype with measurable results.

Key results from the prototype: evolutionary search found severe collapse paths about (1.90\times) better than random search the adversarial method won 19 out of 20 benchmark seeds robustness testing ran across 64 perturbed worlds sensitivity analysis showed dependency strength dominated collapse behavior IEEE-14 physical grid simulation produced a 259 MW unserved-load cascade the system produced uncertainty-aware intervention frontiers instead of a single fragile answer Most importantly, SwanForge makes an invisible risk visible. What we learned We learned that resilience is not only about strengthening individual systems. It is about understanding the dependencies between them. A hospital with backup power may still fail if fuel authorization, road access, telecom, or payment systems break first. A city can look stable at the component level while being fragile at the dependency level. We also learned that adversarial search is a powerful way to reason about infrastructure. Instead of asking, “What failure do we expect?”, SwanForge asks, “What failure would hurt us most if reality were adversarial?”

What's next for SwanForge Next, we want to move SwanForge from prototype to real-world resilience infrastructure. The roadmap is: integrate real municipal GIS and infrastructure data add probabilistic calibration from historical incidents support live digital twins for cities, campuses, ports, and hospitals expand the physical models for power, water, traffic, and communications build expert review workflows for emergency managers and infrastructure planners turn intervention recommendations into operational resilience plans The long-term vision is a new kind of civic safety system: one that discovers black-swan failure paths before the world does.