What problem does it solve? When no dataset can answer a question and judgment is the best evidence available, unstructured group discussion lets seniority and the loudest voice set the number. This Skill runs the RAND Delphi method — anonymous panelists, iterated rounds, controlled feedback of the group median and interquartile range — so expert estimates converge on merit, or the disagreement is honestly mapped instead of averaged away. ## Core Features & Use Cases - Structured panel elicitation: Defines a scoped scalar question, selects a heterogeneous panel of 7–30 experts, and runs blind round-1 elicitation followed by anonymous feedback of the median, IQR, and outlier rationales. - Stability-based stopping rule: Stops rounds when fewer than 15% of panelists revise (Scheibe, Skutsch & Schofer, 1975), never on forced unanimity, and refuses to average a bimodal split. - Deterministic companion tool: scripts/delphi.py (Python 3.9+, standard library only) computes median, Q1/Q3, IQR, outlier flags, inter-round stability verdicts, and Kendall's W for ranked items, with a hand-verified self-test. - Use Case: A leadership team needs a defensible date for when fault-tolerant quantum computing reaches commercial advantage. Convene an eight-expert panel, run three anonymous rounds, and report the 2035.5 median with a 2034.5–2038.5 band plus the unanswered dissenting argument. ## Quick Start Use the delphi-method skill to run an anonymous multi-round expert panel on when our product category reaches maturity, and report the median estimate with its interquartile range and the strongest dissenting view.
