Precise questions only
Every scored forecast must state an observable event, target population or product, verification sources and a clear deadline before publication. Vague claims are not eligible for a verdict.
CORTIST / METHODSHow we verify
How we verify
AI claims and forecasts.
Our rules are public before we start scoring. The goal is accountability, not a perfect-looking track record.
METHOD / EDITORIAL PROCESS · This page describes source attribution, evidence types and corrections. For forecast scoring rules use the protocol; for reviewer blockers use the review dossier; for published results use the scorecard. Editorial responsibility ↗
V1.1 · OCT 22, 2026
How Cortist verifies.
This is our initial editorial standard, not a claim that we have already built a statistically significant forecast record.
Originals are never silently rewritten
We preserve the originally published text, probability, deadline and resolution criteria. Corrections and updates are separately dated, explained and linked to the original. An archival log and independent timestamp verification are part of the planned data infrastructure; this site launch alone does not provide tamper-proof storage.
Separate forecasts from company claims
Cortist forecasts assign explicit probabilities. Company statements are recorded verbatim or accurately paraphrased with primary sources; they are not automatically assigned probabilities or portrayed as contractual promises.
Evidence before verdicts
Resolution requires a stated test and dated, independently checkable evidence. Ambiguous or insufficiently sourced outcomes remain pending or unresolved. News reports are supporting context, not automatic proof.
Show mistakes and uncertainty
Resolved binary forecasts can use the Brier score: (published probability as a decimal − actual outcome, 0 or 1)². Lower is better. We will display question counts, selection limitations and calibration only after sufficient outcomes exist.
Editorial independence
We do not sell favorable forecasts, outcomes or rankings. Any future paid community submission will purchase a defined review/publication service, not a guaranteed result or editorial endorsement. Clear rejection and refund rules will be published before payments launch.
Corrections, disputes and privacy
Readers can challenge factual errors via our contact page. We review disputed sources, document material corrections and protect private submissions. Unverified personal claims and confidential information will not be published.
Topic hubs are living records
Our AGI and superintelligence hubs bring together dated editorial forecasts, sourced third-party statements, research and relevant news. A changing hub is not the same as a fixed original forecast: each locked record is preserved separately, with its own deadline and resolution rule.
Daily monitoring is discovery, not proof
Our research watch checks for newly published or materially revised expert forecasts. Automated discovery is a starting point; human editorial verification checks the original source, timing, definitions and whether a claim is genuinely new. We do not turn vague remarks into invented probabilities or issue automatic verdicts.
AGI and ASI candidates are not scored
F-001 (AGI by 2030) and F-002 (superintelligence by 2040) are preliminary editorial estimates. Their numerical judgments are provisional until objective resolution tests and immutable record procedures are agreed and published. No accuracy or calibration claim is made from these drafts.
Small beginnings. Verifiable progress.
The current forecast record includes eight unscored numeric topic drafts and two separately published, editorially locked milestone forecasts (CM-003 and CM-004). Neither official forecast has reached an independently verified outcome. No success rate or calibration claim is possible. This standard establishes what a Cortist record must contain.
Go to forecast log ↗CORTIST / MODEL v0.1 · 22 OCT 2026
How the year sliders work.
Subjective editorial scenarios—not measured probabilities, a statistical fit, or a consensus forecast.
The slider gives the cumulative chance by 31 December of the selected year, not the chance that the event happens during that specific year. These are Cortist's working checkpoints; values between checkpoints use linear interpolation and whole-number rounding, not independent predictions for every year.
| By year-end | AGI | ASI |
|---|---|---|
| 2026 | 1% | 0% |
| 2030 | 38% | 3% |
| 2035 | 62% | 14% |
| 2040 | 76% | 24% |
| 2050 | 88% | 51% |
| 2060 | 93% | 73% |
| 2070 | 96% | 84% |
ASI is modeled as the stricter threshold; its probabilities never exceed those for AGI in the displayed years. Probabilities stay below 100%, reflecting substantial long-run uncertainty. These checkpoint values are editorial judgments, not taken from the references below.
For external context with different definitions: Metaculus general AI forecast, AI Impacts 2023 researcher survey (HLMI), and Google DeepMind: From AGI to ASI (2026). None of these sources claims Cortist's numerical probabilities.
Version history: On 22 October 2026 the first unscored launch drafts showed 45% AGI by 2030 and 30% ASI by 2040. The same-day model v0.1 replaces these with 38% and 24% at those dates. These are still preliminary; objective outcome criteria and an immutable lock are being developed. No accuracy score is claimed.
This curve is an illustrative probability model, not independently calibrated or validated. All candidate forecasts ↗
OPEN STANDARD / OCT 22, 2026
Forecast Protocol v1.0.
The public technical rulebook and machine-readable record format complement our editorial methodology.
Versioned forecast records are available. Editorial publication locks are now active for CM-003 and CM-004 (27 October 2026); earlier AGI and ASI drafts remain unlocked.
The AGI and ASI candidates have individually published v0.1 JSON snapshots with checksums. Both are drafts. Every additional official forecast still requires fixed YES/NO criteria, an accountable editorial approval, a dated public original and a correction policy. Independent review is optional and must never be implied when absent. Read the protocol · See public results · Download the record schema · Independent review dossier.