AI Executive Office — CXO Assistant Platform · View 17 of 30 · 5 · Runtime
The rule
- The language model never produces a figure, a forecast or a risk score. It classifies the question, frames the scenario and narrates what a deterministic component computed
- This is the single decision that most affects executive trust, and it is also the largest cost control: routing sends most questions down a path with one small-model call rather than a reasoning-model chain
- Where a cell is empty, that component is not consulted. The emptiness is the design
Why it matters commercially
- A wrong narrative is embarrassing; a wrong number is a resignation. Separating them means the platform's failure modes are recoverable
- Deterministic paths are cacheable, testable against expected values, and explainable without reference to a model version
Risks
- The boundary blurs under pressure. "Just let the model estimate it when the measure is missing" is the change request that will arrive in month three, and the answer must be no — the measure gets built, or the platform abstains