How public prediction claims should be audited
Historical data. Clear assumptions. Reproducible evidence.
Start before the result
A useful audit retains the exact public URL, named market, source-assigned date, identifiable selections and a preserved capture demonstrating what was visible before the result. An old upload timestamp alone does not establish what its title or description said at that time.
Keep the entire record
Count every distinct selection and preserve misses as well as hits. Ambiguous and inaccessible claims stay separate from scoreable claims. A multi-number selection cannot be represented as one precise prediction.
Explain the limits
Disclose exactly what content was accessed. A snippet is not a video review. A failed claim or an incomplete record is not proof of fraud. Until the necessary record exists, use Not Independently Verifiable.