Nine predictions with dates on them. We score ourselves in public.

Open

Everything currently on the record

Open predictions, ordered by review date
Ref	Review	Prediction	Reading	Status
002-P3	31 Dec 2027	The AI Futures Project will publish at least one further scenario or timelines update revising its default arrival date, in either direction.	002	Open
002-P4	31 Jan 2028	The organisation will not register as a lobbying entity, form a political action committee, or establish a 501(c)(4) affiliate.	002	Open
001-P4	31 Dec 2028	AI governance will not rank as the single most important issue in any major national pre-election poll during the 2028 US presidential cycle.	001	Open
001-P2	31 Jan 2029	At least one G7 government will have taken an equity stake in, or formal directive control over, a frontier AI laboratory.	001	Open
002-P2	31 Jan 2029	No official US or Chinese government document will adopt "Plan A", "total research transparency" or "mutually assured compute destruction" as a named policy framework.	002	Open
001-P5	30 Jun 2029	Annual global AI datacentre capital expenditure will not exceed twice the US defence budget in any calendar year to this date.	001	Open
001-P1	31 Jan 2030	No binding US–China agreement covering frontier AI training runs, with any third-party verification provision, will be in force.	001	Open
002-P1	31 Jan 2030	The AI Futures Project will still hold no stated position on any social or cultural question — the CU cell will remain unscoreable rather than settling on the axis.	002	Open
001-P3	31 Jan 2031	No OECD member will have legislated a universal unconditional payment, financed from AI or automation revenues, above 25% of national median income.	001	Open

Resolved

Nothing yet

The first review date is 31 December 2027. Until then this section is empty, and it says so rather than being hidden. When entries appear here they will include the ones we got wrong, stated as plainly as the ones we got right, with a note on why the reasoning failed rather than a reclassification of the question.

Where we already expect trouble

001-P2 is the prediction we would bet against ourselves on. Formal directive control over a frontier laboratory is a large, visible act — but partial or informal control through security clearances, classified compartments or procurement dependency could arrive without satisfying the wording. If that happens the entry will be logged as technically correct and substantively wrong, and the wording criticised rather than defended.

Scoring policy

How an entry gets resolved

Resolution happens on the stated date, not before and not when convenient. A prediction that looks likely to fail is not quietly retired ahead of its review.
Three outcomes only: correct, wrong, unresolvable. There is no partial credit and no near-miss category. A prediction that turns out to have been badly worded is logged wrong or unresolvable, never softened.
Unresolvable requires a stated reason. Usually that the wording admitted more than one reading, or that the evidence needed is not public. An unresolvable entry counts as a failure of drafting and is recorded as such.
Resolutions carry their evidence. Each resolved entry links to the public record used to settle it.
The tally is never reset. It runs from the first reading onward, permanently, across every subject.

The reason for all of this is simple, and it is worth stating even though it sounds like a slogan. Being publicly wrong on schedule buys more credibility than being vaguely right forever. Every review date is also a reason for a reader to return, which is a happy coincidence rather than the purpose.