forecasting-ch08-delphilisted

# Chapter 8: The Delphi Room ## Scope and intake Use when structured anonymous expert rounds are needed for sparse-evidence quantity forecasts. Do not use for using convergence of opinions as evidence of accuracy or treating AI personas as independent experts. Ask only for unresolved material inputs: What question and common units will every expert answer? What independent evidence does each have? How many rounds are justified? When will actual outcomes resolve? ## Input contract and additional evidence CSV question,expert,round,estimate,actual; one estimate per question/expert/round. round is a positive integer, estimates share units, actual is finite for CLI scoring and identical across rows for the same question; retain unresolved blank actuals separately. Keep rationale and evidence in a linked qualitative record. Minimal **format illustration**, not sufficient training data: ```csv question,expert,round,estimate,actual Q1,E1,1,100,110 Q1,E2,1,120,110 ``` ## Executable interface Exact CLI columns: `question,expert,round,estimate,actual`. Supported method controls: `as_of, frequency, horizon, season, seed`. The adapter requires complete resolved actuals and compares round medians. Keep unresolved elicitation records in a separate journal until resolution. It does not solicit experts, authenticate independence or estimate a full opinion distribution. ## Applied procedure 1. Write one unambiguous question and recruit experts with distinct relevant evidence. Recor