Forecast Desk
Questions with a deadline, a written resolution rule, and a score that stands.
Every question below is graded by the same collectors that publish the dashboard: a policy rate print, an exchange fixing, a ministry announcement. The rule is written before the forecast, the resolving document is cited, and the record is public whether it flatters us or not.
Open questions · 未决问题
Track record · 战绩
7 resolved · mean Brier 0.024
Calibration is not yet meaningful: 7 resolved questions against a 50 minimum. Scores are published anyway.
Open the scoreboardAgent surface · 接口
The board, each public question and the scoreboard are published as JSON for agents and downstream pipelines.
How the Desk works · 方法
- A question ships only with its resolver. Every question names the feed, the field and the test that will grade it, in the criterion published beside it.
- The archive answers before the model does. A base rate is computed from our own collected history. Where the archive is too young, the packet says so and no prior is invented.
- Independent panellists, median aggregation. Members reason separately on one bounded, cited evidence packet. The disagreement between them is published, not averaged away.
- Resolution is deterministic and never a model's opinion. An event question resolves YES only when the resolving document can be cited.
- Nothing is rewritten. Probabilities are appended; a resolved question is closed. Every score, including the bad ones, stays on the record.
Boundary · 权限
Public questions carry their full probability, criterion and base rate. The rest of the board, the reasoning chains, the evidence packets, probability history and move alerts are part of a Professional membership.
See membership