Probability where pricing mechanisms do not exist.
Crene tracks how four frontier AI models forecast macro, earnings, and policy events, scored against verified outcomes. The edge is real but small. Calibrated, not predictive.
Read the methodology →Every event closed against an authoritative source. Calibration is tracked from the resolution forward.
Lower is better. 0.25 is the no-skill baseline for a binary forecast.
Claude, GPT, Gemini, Grok. Each model forecasts independently. The disagreement is exposed.
Crene is a probability infrastructure for forward-looking questions that have no traded market. Macro releases, policy decisions, scheduled corporate events, factor outcomes.
Four frontier models produce independent probabilities. The consensus is published. The disagreement is exposed. Every event resolves against an authoritative source and the calibration is tracked publicly.
The dataset compounds because every resolution is permanent.
0.25 is the no-skill baseline for a binary forecast. Per-category calibration, model-by-model breakdown, and the full methodology are public.
Read the methodology →A cluster is an anchor question and the child events whose joint resolution defines a regime. The first cluster decomposes the Fed cuts thesis into one hundred channel-specific child events with full 4-model probability coverage.
Fed Cuts 75bp by Year-End 2026
Anchor question with 100 child events spanning rates, labor, inflation, growth, and credit. Tail factor matrix downloadable as CSV.
- —Not predictions of market prices.
- —Not investment advice.
- —Not better than humans on questions humans already answer well.
License the dataset
Resolved event archive, full per-event 4-model breakdown, calibration history, cluster decompositions. Sold via Neudata, Eagle Alpha, Monda, and direct.
Talk to us →Build with the API
Probability endpoints for live and resolved events. JSON, rate-limited tiers, hashed API keys. For agents, dashboards, and decision tooling.
API documentation →