Track record.

Every probability VERA publishes is timestamped before the outcome and scored after it — including the ones we get wrong. This page is deliberately empty of scores today. The first measured Brier scores land here in August 2026, when the first prediction cohort reaches its horizon — and not before.

What exists today

176questions under management 32adjudication rulings to date 176/176questions naming their resolution evidence 361records walked, first falsifier sweep AUG 2026first measured scoring

A question is one situation, one outcome definition, one horizon — held stable across every analysis that touches it, so revisions are visible and scoreable rather than silent. 176 are currently under management, and every one names where its resolution evidence will come from.

How a question resolves

Criteria are pre-registered at publication: what resolves YES, what resolves NO, what proves the analysis wrong.
The question reaches its stated horizon.
Evidence is gathered from cited sources — institutional documents, wire records, the datasets the prior was counted over.
A human adjudicates against the criteria and the cited evidence; the ruling is recorded.
The score is computed and published here — hits and misses at equal prominence.

The discipline before the scores

190+ counted reference classes anchor every probability; the daily falsifier sweep surfaced 69 invalidation candidates on its first run.
Timestamped, unmodifiable records. Every analysis is written once and never edited — the Iran ceasefire call carries a file birth-time 21 minutes before the first public announcement.
Pre-registered resolution criteria. Every scoreable prediction states, at publication, exactly what would resolve it YES, NO, or prove it wrong. The first cohort’s criteria were adversarially tested, repaired, and re-blessed before any score is produced under them.
Epoch stamps. Every scoring-relevant change to the system lands at a stamped boundary. The track record cannot silently rewrite its own past.
Question threading. Consecutive analyses of the same situation share a durable question identity — revisions are visible and scoreable, never silent.

August 2026 — what lands here

Measured Brier scores for the first resolved cohort, against stated baselines.
Every miss published with the same prominence as every hit — the falsifiability register already writes the counter-case for each analysis before publication.

Write to us to be notified when the first cohort scores.

Why publish an empty scoreboard? Because the alternative — claiming accuracy without measured scores — is what this product exists to replace. The machinery that will fill this page is already running in public; the numbers arrive when the horizon does, not when marketing wants them.

site updated 18 July 2026