Methodology · layer
Question design, forecast format, aggregation, scoring and resolution for the probability trackers — merged from the former CitySignals methodology.
Process
Trackers ask precise, resolvable questions and express answers as probabilities with confidence — not pundit commentary. This makes forecasts comparable, scoreable and accountable over time.
Each question declares its type (binary, multiple choice, numeric threshold, range), an official resolution source, open/close/resolve dates, and tags. Questions are curated for relevance and clean resolvability.
A submission carries a probability (or range), an optional confidence, an optional rationale, and a forecast window (early/final/late) and source (public/professional/demo/imported).
Consensus starts as a simple average and moves to a weighted average once forecasters build a track record. Aggregate snapshots are stored over time with participant counts.
Binary/choice questions use Brier-style scoring; numeric questions use absolute error. Scores resolve against the official source and feed public accuracy records.
Where available, forecasts are compared against naive, consensus, market-implied or last-known baselines so skill is visible, not just direction.
Private scenario tools separate factual inputs, assumptions, contextual probabilities and risk flags. They produce ranges and questions for further research — never personalised recommendations.
