On February 28, 2026, the United States and Israel launched strikes against Iran. Within days, Iran retaliated by closing the Strait of Hormuz. The International Energy Agency would later call it "the greatest threat to global energy security in history." Crude flows through the Strait collapsed from 20 million barrels per day to just over 2 million.
Cascade was already running.
What followed was 23 days of continuous, autonomous intelligence production. No human analysts. No editorial meetings. No overnight lag between an event and a briefing. Just a propagation engine tracking how economic stress moves through interconnected systems in real time, producing classified-format strategic assessments at sub-five-minute intervals.
By the time the IMF published its April 14 World Economic Outlook confirming the severity of the crisis, Cascade had already processed over 11,000 analytical steps, monitored 153 entities across 676 connections, and produced predictions that matched or anticipated the IMF's own findings.
This is the story of what happened during those 23 days, what the system got right, what it caught before anyone else, and what it means for the future of geopolitical intelligence.
The Architecture: How Cascade Thinks
Before diving into the results, it helps to understand what Cascade actually is. It is not a dashboard. It is not a news aggregator. It is not a large language model summarizing headlines.
Cascade is a discrete-time propagation engine built on a directed graph. Every entity in the system (countries, sectors, markets, commodities, organizations) is a node carrying a four-dimensional state vector: Stability, Economic, Political, and Social. These four dimensions allow the system to model asymmetric responses. A sanctions event hits an economy differently than it hits political stability, and those differences matter for predicting what happens next.
Edges between nodes represent transmission channels. When crude oil spikes, that shock does not hit every connected entity the same way or at the same time. Each edge carries its own weight, delay, and nonlinear transfer function. Some transmission channels saturate. Some only activate above certain thresholds. Some accelerate as stress increases. The system models all of these dynamics simultaneously.
At each time step, shocks propagate through the network. Entities absorb impact based on their own persistence, sensitivity, and restoration characteristics. The system is mathematically guaranteed to remain stable through spectral radius analysis of the network's companion matrix, meaning it produces useful predictions rather than runaway divergence.
Monte Carlo simulation runs hundreds of independent trajectories with stochastic noise to produce confidence intervals and probability distributions. And a continuous self-calibration engine compares predictions against observed market data, automatically adjusting parameters when confidence thresholds are met.
The result is a system that does something no human analyst team and no existing platform can do: model how a shock in one part of the global system propagates through delayed, nonlinear, multi-dimensional transmission channels to produce second, third, and fourth-order effects across every connected entity. In real time. Continuously.
We backtested Cascade against 9 historical crises before deploying it live. 121 entity predictions. 91% direction accuracy. 86% acceptable grade rate (A or B). From the Russia-Ukraine invasion to the Lehman Brothers collapse to Fukushima to the UK Gilt Crisis. No cherry-picking. Every prediction graded and published, failures alongside successes.
The Iran-Hormuz crisis was different. This was not a backtest. This was live.
Week 1: The Cascade Begins (Steps 1–4,024)
By Step 4,024, roughly seven days into the crisis, Cascade was monitoring 127 entities across approximately 325 connections. The system had already identified the core propagation architecture of the crisis.
The Strait of Hormuz as single-point-of-failure. Cascade flagged the Strait as the primary transmission vector for cascade acceleration before the IEA made its historic characterization. The system traced crude oil's impact through five distinct channels: direct inflation on energy importers, fiscal drain on subsidizing governments, transportation cost escalation across global supply chains, currency depreciation in petroleum-import-dependent economies, and supply chain disruption across manufacturing networks.
The gold inversion. This was perhaps the most striking early finding. Gold showed a negative 49.8% economic impact deviation despite a 20.3% price appreciation. Cascade correctly interpreted this as forced liquidation rather than safe-haven buying. When precious metals are sold into a geopolitical crisis rather than bought, it signals that investors are meeting margin calls and managing currency outflows, not hedging uncertainty. Gold was transmitting systemic risk, not absorbing it.
This is the kind of counter-intuitive, second-order read that separates real intelligence from news aggregation. Most platforms would simply report "gold up, crisis mode." Cascade caught the mechanism underneath.
ASEAN subsidy vulnerability. Cascade identified that Thailand, Singapore, and the Philippines faced subsidy unsustainability within 30 days if crude oil stress persisted. On March 24, the Philippines declared a state of national energy emergency. Cascade's prediction preceded the event.
Week 2–3: The Network Densifies (Steps 4,024–11,139)
Over the next 16 days, the system evolved significantly. Entity count grew 20% from 127 to 153. Connection density more than doubled from approximately 325 to 676 edges. The network added Argentina, the Caribbean, Greenland, Hong Kong, Kashmir, Madagascar, Sri Lanka, Sudan, Vietnam, and over 20 sector-specific entities.
But the quantitative growth masks a qualitative leap. By Step 11,139, Cascade had developed capabilities that were absent at Step 4,024:
Cascade magnitude quantification. The system began producing specific numerical cascade magnitudes. The peak systemic cascade magnitude hit 108.6 at the t+30 projection window. This is not an arbitrary number. It reflects the compounded, multi-path stress accumulation across all six transmission channels (energy prices, supply chain logistics, currency depreciation, financial volatility, inflation dynamics, and policy constraint) normalized against historical baselines.
Probability-weighted systemic failure thresholds. Cascade calculated that if policy coordination failed or the Strait closure extended beyond 14 days, the probability of systemic failure (30%+ deviation across the network) would exceed 50% by t+72. That is a specific, falsifiable prediction with a defined probability and timeline.
IMF-calibrated scenario modeling. The Step 11,139 report directly referenced and aligned with the IMF's three-tier scenario framework: the reference forecast (19% energy price increase, 3.1% global growth, 4.4% headline inflation), the adverse scenario (sharper price increase, 2.5% growth, 5.4% inflation), and the severe scenario (extended disruption, 2% growth, 6%+ inflation). The IMF published these figures on April 14. Cascade was operating within the same quantitative framework before that publication.
What Cascade Got Right: The Validation Record
The Iran-Hormuz crisis has become a live benchmark. Here is what Cascade predicted and what subsequently happened:
Strait of Hormuz as existential energy risk. Cascade flagged this at Step 4,024. The IEA Executive Director subsequently called it "the greatest threat to global energy security in history." Crude flows collapsed from 20 mb/d to 2 mb/d. One billion barrels of production expected lost.
UK as most vulnerable developed economy. Cascade assigned the UK the highest deviation score among developed nations at 7.0%, with a 10.1 percentage point decline in political stability. Subsequently, UK inflation was broadly projected to breach 5% in 2026, and the ECB postponed planned rate reductions. Multiple independent analyses confirmed the UK as the worst-hit major economy.
ASEAN energy emergency. Cascade predicted subsidy unsustainability across energy-dependent ASEAN economies. The Philippines declared a national energy emergency on March 24. Vietnam abolished fuel levies. Bangladesh began facing recession-like conditions. Pakistan, Bangladesh, and Vietnam were identified among the worst-hit Asian economies.
Afghanistan humanitarian tipping point. Cascade identified Afghanistan as approaching systemic collapse with nine in ten families in food insecurity. UNDP subsequently confirmed this assessment, reporting average monthly household income at $99.76, tripled rents from the 4.5 million returnee surge, and women's workforce participation at 6%.
Fertilizer and food security cascade. Cascade identified that fertilizer supply disruption through the Hormuz chokepoint would cascade into agricultural costs during Northern Hemisphere planting season. This was confirmed: over 30% of global urea exports transit the Strait, and the Food Policy Institute warned of long-term food price increases.
Energy-driven inflation trap for central banks. Cascade identified the impossible policy bind: central banks cannot simultaneously manage energy-driven inflation and support growth without triggering currency depreciation. The IMF subsequently stated that the crisis "poses immediate policy trade-offs: between fighting inflation and preserving growth."
Pakistan remittance vulnerability. Cascade identified Pakistan's current-account deficit as dependent on Middle East remittances that would contract 20-30% in a Gulf recession. Pakistan was subsequently confirmed among the worst-hit economies, with severe fuel shortages and IMF growth downgrades.
Ten out of ten major analytical claims confirmed by institutional sources. Four predictions made before the confirming events occurred.
What the Competitive Landscape Looks Like
The geopolitical risk intelligence market is built on three models, all of which have fundamental limitations that the Iran crisis exposed.
The advisory model (Eurasia Group, EY Geostrategic Business Group) employs hundreds of human analysts across dozens of offices. They produce deeply informed, nuanced assessments. Annual contracts run $150K to $500K+. Their limitation: latency. When the Strait of Hormuz closes, a human analyst team requires hours to days to produce a written assessment. During those hours, the cascade is already propagating through five transmission channels. The analysis arrives after the second-order effects have already begun.
The platform model (Stratfor/RANE, Recorded Future, Dragonfly) provides real-time alerts and daily intelligence briefs. Pricing ranges from $40K to $250K annually. Their limitation: they monitor events, not propagation. They can tell you the Strait of Hormuz closed. They cannot tell you that the gold market is transmitting systemic risk rather than hedging it, or that crude oil's economic impact is propagating through five distinct channels with different delays and nonlinear dynamics in each.
The data model (S&P Global, Stabilarity) provides quantified country risk scores and continuous data feeds. Their limitation: they score entities independently. They do not model how a shock to one entity cascades through weighted, delayed, nonlinear transmission channels to affect connected entities. They produce snapshots, not propagation maps.
None of these models do what Cascade does. None of them model directed, delayed, nonlinear cascade propagation across a multi-dimensional state space with continuous self-calibration against observed data. None of them produce the kind of counter-intuitive, second-order signals (gold as risk transmitter, structural versus headline GDP disaggregation, synchronized cross-asset liquidation patterns) that define the difference between intelligence and information.
The combined annual cost to approximate Cascade's capabilities through existing providers would exceed $400K, and the result would still lack the propagation engine.
Known Limitations: Where We Fall Short
We published our limitations alongside our backtesting results because we believe transparency builds trust faster than marketing.
Systemic financial crises. In our backtesting, both the Lehman Brothers collapse and COVID-19 scored 73% at the A+B level. The model is calibrated for shocks in the 1 to 2 sigma range. When credit-driven cascading failures produce 3+ sigma movements (VIX at +486% during COVID), our magnitude estimates are too conservative. We capture direction correctly but underestimate the extremes.
VIX magnitude. We consistently underestimate VIX peak levels. The VIX has convex, nonlinear behavior that our current transfer functions capture directionally but not at full magnitude. This is a known, documented weakness.
Currency cross-rates. EUR/GBP and USD/EUR have been direction misses in backtesting. Cross-rates depend on relative dynamics between two economies, and our current architecture treats each currency entity's state independently.
Source depth. The current system ingests 12 intelligence sources per assessment cycle. For production-grade deployment, this should scale to 50 to 100+ sources. The analytical framework is strong, but source breadth is a scaling challenge we are actively addressing.
Recovery timing. The model shows sustained drawdowns correctly, but exact recovery day predictions remain noisy due to stochastic dynamics and the inherent unpredictability of diplomatic and policy interventions.
We are not claiming perfection. We are claiming that a directed propagation engine with 91% direction accuracy across 121 historical predictions, now validated against the largest energy crisis in modern history, represents a fundamentally different approach to geopolitical intelligence.
What Comes Next: The Digital Operator Convergence
Cascade produces intelligence. But intelligence without action is just reading.
The next chapter is the convergence of Cascade with MetisOS Digital Operators: persistent, autonomous AI agents that carry organizational context in long-term memory. A Digital Operator embedded inside a firm does not just receive a Cascade briefing. It reads the briefing, maps it against the firm's specific vendor relationships, client exposure, contractual obligations, and supply chain dependencies, and then acts. It drafts the supplier communication. Adjusts the project timeline. Notifies the relevant team. Updates the client.
Consider a consulting firm with ASEAN-based clients. Cascade identifies that the Philippines and Thailand face subsidy unsustainability within 30 days. A Digital Operator with organizational context knows which clients have exposure to those markets, which engagements have deliverables affected by those timelines, and which partners need to be briefed. It does not wait for a Monday morning email. It acts in real time, within the firm's own voice and protocols.
That convergence, between real-time cascading risk intelligence and persistent autonomous agents with organizational memory, does not exist anywhere in the market at any price point. It is the architecture for what we believe organizations will need in an era of accelerating, interconnected, systemic risk: not just awareness, but autonomous resilience.
The 23-Day Record
The numbers:
- 11,139 processing steps completed
- 153 entities monitored across countries, sectors, markets, and commodities
- 676 directed connections tracked
- 100% peak severity event processed (Iran-US supply chain disruption)
- 10/10 major predictions subsequently confirmed by IMF, IEA, UNDP, and market data
- 4 predictions made before confirming events occurred
- 0 human analysts involved in production
- 23 days of continuous, autonomous operation
The Strait of Hormuz is still constrained. The crisis is still unfolding. Cascade is still running.
If you want to see how it works on live data, you can request a demo. If you want to see how we performed against history, the full backtesting results are published at /benchmarks.
Christian Johnson is the Founder and CEO of Metis Analytics, the company behind Cascade and MetisOS. He can be reached on LinkedIn or at cjohnson@metisos.co.