Push the button. The model runs the world two hundred times under that scenario. The cone is the answer.
You see what already happened. You feel briefly informed. You scroll on, none the wiser about what tomorrow does to the position you're holding.
This is the news with nicer plots. We do the other thing.
The library has the obvious ones. Fed cuts 50. Oil to $120. 10y doubles in a quarter. GES lets you stack interventions and run them at the same time.
You're not picking a stock. You're picking a question.
A GAM learns the activation shape of every channel. Regime detection keeps state-aware noise. Then a stochastic forward simulator runs the system two hundred times through your conditioning, with empirical residuals for the parts the model doesn't pretend to understand.
Each path is a different roll of the dice through the same physics.
The 50% band is the model's working honest opinion. The 95% band is the room it's leaving for being wrong. Wide cone means the system is a coin flip under your scenario. Narrow cone means it has somewhere it tends to go. Lopsided cone means it has somewhere it tends to go fast.
The point is to make uncertainty legible, not delete it.
Regime clustering on minute-tape windows finds the system's resting states — easing, drift, risk-on, risk-off. The cone fades through them at different speeds in different colors.
Where the bands switch is where the math thinks the world's mood does.
March 12. The model said SPY +0.4% to +1.6% on the 25 bp cut. Realized: +1.1%. In the 50% band. We log all of them. Three tail misses out of thirty-eight calls in the last ninety days. We tell you which.
The default future isn't bullshit because we keep score.
The cone is honest about its uncertainty. So is the log. If the model misses, it's on this page. We don't take it down.