Resolution Day, Saturday, Oct 31: 5 AI forecasts graded in public

Each of these was published days or weeks ago as a dated yes-or-no question with the index's odds. On Saturday, Oct 31 each is graded in public, and the misses get the same space as the hits. This page updates at every refresh (5:17 AM, 12:17 PM and 6:17 PM ET).

The score so far

1 of 5 graded: 0 right, 0 wrong, 1 toss-up (not counted).

Will OpenAI publish its 21st report of its own AI misbehaving, by Oct 31?

Open The index says likely (70%). Graded when the evidence lands, right or wrong.

What would settle it: The log itself; OpenAI safety posts referencing a new incident.

Will a court approve the rules for selling Character.AI, with Disney's $700,000 break-up fee intact, by Oct 31?

Open The index says likely (60%). Graded when the evidence lands, right or wrong.

What would settle it: The Oct 30 10 a.m. ET hearing docket; a revised bidding-procedures order; Disney amending its terms before the hearing.

Will Britain say all 32 government services it switched off are back, by Oct 31?

Open The index says a toss-up (50%). Graded when the evidence lands, right or wrong.

What would settle it: GOV.UK service status pages; NCSC statements; the Cabinet Office's briefing to Parliament.

Will America's cyber agency formally warn that the flaw under most websites is being used in attacks, by Oct 31?

TOSS-UP, NOT COUNTED Answer: yes, graded Oct 26, 2026. The index had said a toss-up (50%).

CISA added CVE-2026-109618 and CVE-2026-109619 to its Known Exploited Vulnerabilities catalog on Oct 26 (version 2026.10.26, released 22:24 UTC), confirmed by opening the feed directly on Oct 27. The evidence

Will America's cyber agency order federal agencies to act on the flaw under most websites, by Oct 31?

Open The index says very unlikely (15%). Graded when the evidence lands, right or wrong.

What would settle it: cisa.gov/news-events/directives; a federal agency disclosing a PHP2Shell intrusion.

How a call is graded: the question's exact wording, fixed when it was published, decides it; a call at exactly 50% is graded but not counted. Every call page shows the fingerprint of its wording and the commit that added it. The method · every forecast and the track record.