Explore the public record.

1. 1
   Collect and verify

   Two passes confirm the quote, publication date, speaker, future-facing claim, and high-confidence attribution.

2. 2
   Define the test first

   Every new call gets an inclusive ISO resolve-by date and written criteria before outcome research. Open-ended claims are excluded.

3. 3
   Match the resolver

   Official records and attributable reporting resolve each call under its frozen rule.

4. 4
   Guard the timeline

   A contingent call stays private and counts nowhere until its antecedent is separately verified as satisfied with dated evidence; then its consequent may appear as pending.

5. 5
   Separate close calls from timing misses

   A NO earns half credit only for a preregistered numeric threshold missed by at most 5%, or a deadline missed by at most seven days. When the same event occurs farther outside the explicit window, it may be labeled an early or late timing miss but remains zero points.

6. 6
   Consolidate repetitions and measure boldness

   The same fundamental claim by the same pundit counts once, with every sourced occurrence preserved on its receipt. Exact repetitions group automatically under strict speaker, deadline, condition, polarity, and resolution-rule guards; paraphrases require an approved review group. Two independently sourced episodes set the one canonical prediction to 100 boldness, even when the wording is identical. Every occurrence retains its own source provenance.

7. 7
   Publish the evidence

   The public record is researched and linked. Near-miss and timing-miss reasons, polling context, and corrections remain visible.

Accuracy = (YES + 0.5 × eligible near misses) ÷ resolved YES/NO predictions. The underlying NO verdict never changes. Timing misses remain zero-point NO results. They identify the same event and actor occurring outside an explicit window; wrong actors, different events, and vague calls remain full misses.

Boldness never affects accuracy. Polling context is never applied retroactively. Polling is context for an election-winner call, not its outcome: certified results still resolve the prediction. A poll-specific prediction may use polling as researched evidence after review. If no verified, recent pre-publication snapshot exists, the score uses statement structure alone unless the independently repeated-call rule applies; the receipt says which basis was used.