CADi Decision Engine
How it decides. How it doubts itself. How it refuses to act when it can’t be trusted.
Every contact flows through these nine steps in order. CADi scores it through four analytical lenses, measures how much they agree, weighs candidate decisions against that uncertainty, applies safeguards, and refuses to automate when the decision is fragile. Each step maps to real code, and every decision is written to a tamper-evident audit log.
The four lenses intentionally disagree — their disagreement is what the engine measures as uncertainty.
This is a prototype. The pipeline runs and every step is auditable, but the scoring weights are hand-tuned, the causal graph is assumed, and nothing here has been validated against real outcomes yet.
CADidoesn’t pretend to be smarter than it is. That’s the point.