Start a task with public reference data or a local CSV/JSON file. Run data checks and descriptive calculations, inspect the execution record and download the results. Language-model reasoning is a separate connection.
A specified research question, evidence cutoff, source packet and review criteria.
Work
Planned independent submissions, comparison of assumptions and a separate challenge round.
Take away
Planned model-by-model rationale, uncertainty, abstentions, costs and preserved revisions.
A separate three-stage, single-provider research-drafting adapter is implemented in the workspace. A live model response has not been verified; that sequence is not an independent cross-model benchmark.
Forecast Observatory
Public preview · snapshots and scoring
Let the outcome test the judgment.
A public starting point for questions, source records and an understandable evaluation rule.
A rule before a forecast. A result before a score.
Bring
Dated official registry references and question drafts that still require protocol review.
Work
Search and filter references, inspect a draft, download its source and explore Brier scoring.
Take away
Unsubmitted question drafts, published source snapshots and local scoring simulations.
Dedicated forecast freezing, independent time attestation and prospective outcome adjudication remain planned. No validated forecasting track record is claimed.
DESIGNED FOR CLINICAL RESEARCH
Keep the scientific question in view.
The product direction includes clinical strategy research, trial-design scenarios and prospective readout questions. Each needs explicit assumptions, qualified review and a clear distinction between a proposed study and an observed result.