Retained model estimates
The historical archive includes probability estimates attributed to Claude, OpenAI, Gemini and Grok models. Model versions and observation dates vary across records.
Inspect retained model probabilities and resolved outcomes from the historical public corpus. Public example updates have been retired; these figures do not measure current private workspace forecasts.
Dated estimates. Recorded outcomes. Inspectable history.
Loading the reported corpus summary.
The broad corpus is separate from the timestamp filtered snapshot benchmark and the macro only subset. Its legacy scores do not establish current model skill, private workspace performance or investment returns.
The headline outcome count is the full Crene native set reported by the API. Family rows combine model versions and count stored scored estimates, so samples may differ from the consensus population. The reported identifier is not a version-specific evaluation cohort.
The directional accuracy numerator and denominator will appear when the public statistics are available.
The legacy summary classifies consensus above 50% as Yes and all other values as No. These reported values use the historical endpoint’s population and default rules; they have not been recomputed in this page.
View methodology and benchmark definitions ↗Reported scores for historical model families. Lower Brier means smaller probability error on that sample; different versions and populations prevent a controlled model ranking.
Loading the public record…
Events provide scored outcomes. Thesis, Factor and Scenario records organize the assumptions, distributions and pathways around them.
Loading the public record…
The historical archive includes probability estimates attributed to Claude, OpenAI, Gemini and Grok models. Model versions and observation dates vary across records.
Defined Yes / No questions retain probabilities and recorded outcomes. Inspect each question, horizon, forecast timestamp and resolution evidence before reusing a row.
Saved assumptions and prior states provide context around a view. Public example pages and their automated updates have been retired.
Continuous variables use saved distributions and drivers. Their presence in the corpus does not establish that a fresh assessment is available.
Authored pathways preserve alternative futures and their components. A pathway count describes structural coverage, not an event probability.
A sequence of saved estimates shows how a recorded view evolved. Repeated snapshots and related questions are not independent observations.
The public API reports legacy consensus and model-family summaries. Numeric outcomes and scenario pathways are not given binary Brier scores here.
Personal forecasts, saved model cohorts and reviewed outcomes are measured within their authorized workspace. They are separate from this historical public corpus.
Many questions can still depend on a small set of shared drivers. Question density shows how tightly each record’s components cluster by domain.
Question density is a structural read, not a correlation estimate. Exact effective independence and covariance figures are not published before the observation record can support them.
Loading the public record…
Inspect the question definition, forecast date, recorded outcome and its supporting source together. The historical resolver classifies grounding URLs by source tier; classification is an annotation, not an enforced evidence gate or independent source verification.
Government agencies, central banks, regulators and exchanges, including SEC filings, BLS, BEA and Federal Reserve releases.
Financial news and data providers such as Reuters, Bloomberg and AP. A source classification does not verify the underlying claim.
Secondary sources and some search grounding redirects. These labels annotate provenance; they do not by themselves establish usable resolution evidence.
Private workspace outcome review records an editor’s evidence attestation and a checked forecast cutoff. That review remains distinct from independent verification. Structural records, continuous variables and scenarios need their own evaluation rules.
Explore the sample exports or request data for an internal research workflow.
Public exports expose the historical corpus permitted by each endpoint. Private workspace records and team research are not part of these public read routes.
For a team evaluation, agree the intended use, record coverage, delivery requirements and access scope before relying on an integration.
Review a thesis with Crene ↗