15 publications, 2011 to 2025 · 78 predictions
each graded by two agents who cannot see each other's verdict (D7)
3agreed
agreement beyond chance, kappa 1.00
0settled
the graders disagreed; a third agent decided (D20)
3audited
an agent re-checked a fixed sample of agreed verdicts: 0 corrected, 0 left unsettled (D11)
3published
every claim has a verdict
Measured under its own rule, not as a hit rate: which trends the publisher kept, renamed or dropped (D33). See the series.
Validated: two blind graders, an adjudicator on every disagreement, and an agent audit of a fixed sample (D20).