Gauge

Group estimates go wrong in a well-documented way: the first number spoken anchors every number after it (Tversky & Kahneman, Judgment under Uncertainty, 1974), and disagreeing with the room is socially expensive. The crowd is only wise when its guesses are independent (Surowiecki, The Wisdom of Crowds) — which is why structured methods like Delphi collect estimates before discussion. Gauge runs that loop live: silent submissions, one simultaneous reveal, the spread itself as the artefact. When the room splits, a Delphi second round: discuss what the spread shows, silently re-estimate, and see how far the room converged. If everyone's intervals overlap, commit and move on; if the room splits into camps, that's the discussion — found before anyone anchored it.

Probability questions are answered on a 0–100% slider with verbal anchors; quantity questions as a 90% interval, so people can be honest about uncertainty instead of faking a point estimate. The reveal shows every interval, the zone everyone believes (or its absence), medians, and camps — with one quotable headline per question. A pooled room range is an elicited judgement, not automatically a calibrated forecast or evidence: Fermi can receive it only as a review-needed input to restate or adopt.

Privacy: your questions never leave the link — the server sees only numbers, for 24 hours. The session definition travels in the URL fragment; the relay stores anonymous numbers (plus display names only if you switch names: on) and deletes everything 24 hours after the session's last activity — or immediately, when the facilitator ends the session after the reveal.