The model
9 discrete triggers, weighted by blast radius, feed 6 feedback loops that compound against 5 slow-moving amplifiers. The result is a single composite stress index between 0 and 100. Not a probability. Not a forecast.
Triggers fire
9 discrete events with current marginal probabilities, weighted ×1, ×2, or ×3 by blast radius.
Loops propagate
6 feedback pathways carry stress from one trigger to others — the configuration is more than the sum of its parts.
Amplifiers compound
5 slow-moving state variables (valuation, concentration, CRE wall, IG supply, stablecoin backstop) decide how much pain the system absorbs.
Show the math (live numbers)
| Trigger | Tier | Weight | Prior | p_max | w·p | w·p_max |
|---|---|---|---|---|---|---|
| AI capex shock | T1 | ×3 | 0.30 | 0.50 | 0.90 | 1.50 |
| Oil & Iran / Hormuz | T1 | ×3 | 0.45 | 0.45 | 1.35 | 1.35 |
| Rates and fiscal stress | T1 | ×3 | 0.12 | 0.35 | 0.36 | 1.05 |
| Taiwan | T1 | ×3 | 0.09 | 0.25 | 0.27 | 0.75 |
| Treasury basis trade | T2 | ×2 | 0.12 | 0.30 | 0.24 | 0.60 |
| Private credit dislocation | T2 | ×2 | 0.25 | 0.35 | 0.50 | 0.70 |
| Yen carry unwind | T2 | ×2 | 0.27 | 0.40 | 0.54 | 0.80 |
| AI cyber incident | T3 | ×1 | 0.16 | 0.25 | 0.16 | 0.25 |
| Stablecoin de-peg | T3 | ×1 | 0.05 | 0.20 | 0.05 | 0.20 |
Trigger component: Σ(w·p) / Σ(w·p_max) × 75 = 4.37 / 7.20 × 75 = 45.52
Amplifier component (25% weight, normalized over 5 amplifiers): 15.55
Total: 61/100 · SEVERE
Cross-family validation
Every prior change is validated by two independent LLM families (Claude and GPT-5). Published numbers require both families to agree within a stated tolerance band. All prior changes are written to the auditable forecast log →.