Methodology

The tracker includes public, attributable AI forecasts that can be connected to a date, source, forecaster, and later evidence. A claim is easier to score when it has a concrete time horizon, observable outcome, or falsifiable threshold.

Broad worldview claims, recommendations, and risk arguments may appear only when they are part of the public forecasting record. They are not scored as resolved unless the public evidence is specific enough.

Resolved hit rate is calculated as confirmed predictions divided by confirmed plus incorrect predictions. Still-unfolding and too-early-to-score claims remain visible in the ledger but are excluded from that rate.

Claim confidence describes how directly a public statement is phrased as a forecast in the source material. It is not a probability estimate and does not mean the forecast is likely to be correct.

High-confidence entries are usually explicit forecasts or thresholds. Medium-confidence entries are more interpretive but still traceable to the public record. Implied entries should be read with extra caution because the forecast framing is inferred from broader public claims.

Confirmed and incorrect statuses are paired with a score-strength label. This separates the outcome from how literally the public evidence matches the forecast. It prevents aggregate or proxy-supported confirmations from looking identical to direct confirmations.

Current resolved record: 12 resolved predictions; every resolved prediction has a score-strength label and rationale.

Each prediction card carries source material, references, a last-reviewed date, and an AI-assisted assessment. Primary sources, original forecasts, official releases, technical reports, and public benchmark records are preferred where available.

Reputable reporting and institutional summaries are used when they are the clearest public evidence available.

Statuses can change as new evidence arrives. The ledger is reviewed periodically, not in real time. Scores reflect the public record as of the listed review date. A score is an editorial judgement about the public record as of the listed review date.

Resolved cards use a structured evidence map: each evidence item states what part of the score it supports, how directly it supports it, and which numbered references on the card were used. This is separate from the narrative assessment so readers can inspect the support path rather than only trusting the summary.

This improves auditability but does not replace external source-passage verification. A 9.5-grade launch should still spot-check the cited passages themselves, especially where the support type is aggregate, partial, proxy, or context.

Source-passage verification is the manual review layer above the structured evidence map. Each resolved evidence item has one audit record keyed to the prediction ID, evidence-item number, checked date, verdict, and numbered references that were verified or access-limited.

This audit checks whether attached references support the public evidence item. It does not imply endorsement by the forecasters, exhaustively re-score unresolved claims, replace legal review, or prove that paywalled and bot-blocked pages are independently accessible to every reader.

Source coverage is shown on each prediction card. A linked source is the original forecast material, an outcome reference, or a context reference attached directly to that card. Coverage is not a truth guarantee; it is a disclosure of how much public material is attached to the assessment.

Multi-source review means the card links the original forecast material and at least one additional review reference. Single-source review means the forecast source is linked, but separate outcome/context references are not listed on that card. Missing linked sources are treated as repair issues by the source-health audit.

Material scoring changes should be logged with the review date, affected claim, new status, previous status, and source that changed the record. This release has one public baseline entry:

1. 2026-06-04 · Current scoring baselineCurrent prediction statuses, evidence notes, source links, methodology page, and source-passage checks reflect the active public baseline for this static release.

Resolved predictions have public decision records that list the current status, score strength, review date, and reviewer note. This is not a private audit trail; it is part of how the public ledger explains its scored outcomes.

This index is not a complete database of every AI forecast, and it does not claim to be neutral in the way a formal forecasting tournament would be. It is a transparent editorial tracking project with source links and visible scoring notes.

The tracked forecasters are not affiliated with this site. AI-assisted summaries and assessments may contain mistakes, omissions, or bias. Do not rely on this site for investment, legal, policy, or other consequential decisions without independent verification.

The best correction request identifies the prediction, the disputed status or evidence claim, and a public source that changes the scoring record. Corrections can be sent to scott@scotthazlitt.ai.