The interview
Twelve questions, four rounds, adaptive. It never asks for a number it can look up — only for the things you alone know.
Why it is built this way. A wizard that asks the sponsor for their numbers and then computes a recommendation from those numbers is a spreadsheet wrapped around their own optimism — more dangerous than a spreadsheet, not less. So: ranges rather than point estimates, drivers sampled together rather than independently, gates kept on their own axis instead of blended into a composite score, and beliefs captured before the reference class is revealed.
Reference-class ranges are illustrative seed values for a prototype, not licensed market data — replace them before this informs a real decision. Every driver carries provenance, so the swap is mechanical. Runs are seeded and reproducible: the same inputs always produce the same answer. Nothing here is investment advice.
Build 2026-08-02T22:10:28Z — check this against your deploy to confirm you are running the current version.