Claim classes

Contract 0.1, issued in publication v0.3. Unchanged.

Pages, charts, and forecasts that conflict with this page are wrong. This is the scoring contract. It was written before any forecast was published. A claim that cannot be filed under these rules is not published.

Four classes

Every public claim carries exactly one class. The class is visible next to the claim. It is not implied by tone.

Measured. A number or fact copied from a named external source, with the retrieval date and a stable URL. A measurement is not a forecast. A recalculation is a separate claim, classed Modeled, and it links to the source number it departs from. Required: source, publication or vintage date, retrieval date, the exact quantity, the unit, and the known bias not being corrected.

Extrapolated. A continuation of a published series under a stated fit. Required: the series, the fit window, the functional form, the interval, and at least one break condition. “Trends continue” is not a break condition. “AI will be able to …” without the if-clause is rejected.

Modeled. The output of a named model under a frozen parameter set. Changing a parameter creates a new version. A model output is not, by itself, a prediction that the world will match it.

Argued. A bound, a distinction, or a path claim that no scheduled observation can fully settle. Required: the proposition, the premises, and the evidence that would withdraw or narrow it. Argued claims are not scored. “Superintelligence by 2032” is not Argued. It is a forecast, and it needs a resolution rule or it is not published.

What may be scored

Only dated forecasts are scored. Measured and Argued claims are never wins or losses. A forecast ships only if these are frozen: one-sentence claim no stronger than the resolution block, method class, operational definition, named resolution source, absolute resolve-by date, interval or binary cut, falsifier, version id. Missing any line, it does not ship.

Resolution

Authors do not resolve forecasts. Resolution is a comparison of the frozen block to the named source. The block wins over a stronger headline. A late source is waited on. Silence is not a No. If the source will never publish that quantity, the row is Annulled, not repaired. An annulment is not a hit and not a miss.

A forecast can resolve Yes, No, a numeric value, Annulled, or Superseded. The old version remains readable.

No silent edits

Wording that would change how a forecast resolves is a new version. The old text stays. Spelling and broken links may be fixed without a version bump.

Out of scope

Version 0.1 does not score equity prices, unemployment, extinction or catastrophe probabilities, policy recommendations, or claims that resolve on a person or a lab using a particular phrase. Those topics do not share a confidence mark with a capability measurement.

Filing minimum

A ledger row has: id, claim, class, definition, source, opened, resolve-by, interval, falsifier, version, supersedes, status. Empty scored fields mean the row is a draft.

See also Opening threshold and Domains.