COLDTELL
Signal radar
14 live · 1 archived
upd 26 Aug 2026
A forecast is useful only if its wording, evidence, odds and later result stay attached. This page states what the system can prove today, what it cannot, and how publication works.
Each percentage is attached to an exact YES/NO rule, date and public resolution source. Changed wording or odds create a new immutable call; the superseded call remains visible as a void, never an overwritten record.
3 calls across 3 independent event families have scored, with an event-weighted Brier of 0.18 (0 is perfect, 0.25 is a constant 50% forecast). That sample is too small to mean anything. Coldtell will not claim forecasting skill before at least 30 independent event families resolve.
Sources are graded for authority, independence and incentive. Several articles repeating one press release remain one origin. Every dossier exposes the recorded links, source notes, exact claim and append-only revision history.
15 forecasts have an internal challenge panel covering the outside view, a causal rival, measurement risk and residual risk. Independent external reviews completed: 0. The site never presents internal role simulation as human peer review.
Correlated forecasts are grouped into event families and thesis clusters. Ranking discounts overlap, and calibration gives one shared event total weight one even when several checkpoints use the same release.
Only the generated public directory is deployed. The site has no accounts, forms, cookies, analytics, advertising or sponsored placements. External evidence opens without a referrer. This is research commentary, not financial, investment or medical advice.