StewAI Blog · Decision Science · 15 min read
They Approved 1.2 Million. The Outside View Said 3.3, and Refused to Be Sure.
An approved CHF 1.2 million budget, run through a workflow that ignores the plan on purpose and asks only how projects like this actually turned out. The outside view came back with 3.35 million expected and over six million worst case, then refused to call three data points a confident forecast.
Frequently asked questions
What is reference class forecasting, or the outside view?
Instead of adding up your own plan, you place the project in a distribution of finished comparable projects and read the estimate off the base rate. Kahneman and Lovallo showed the inside view is systematically optimistic; Bent Flyvbjerg turned the outside view into a formal method that was adopted into UK planning guidance.
Why did the workflow refuse to give a confident forecast?
Only three comparables survived independent validation, and the recipe's rule requires five for a confident recommendation. A deterministic gate counted three, routed to a thin-class memo with a THIN CLASS badge, and the confident decision memo never ran. The gate is counting, not judging.
What did the run conclude about the CHF 1.2 million budget?
The surviving class had a mean outcome ratio of 2.79, minimum 1.45, maximum 5.47, so the anchors were 3.35 million expected, 1.74 million optimistic, and 6.56 million conservative. All three comparables came in over plan, and the report labeled the numbers indicative only because the class was thin.
Read the full article on StewAI