Leaderboard / System profile
Claude Sonnet 5.5
By dataset
Index with 95% interval. Grey band: range of all other systems. Dark tick: their median.
Score components
Each score combines these three parts, each from 0 to 1. Red tick: random answers.
By topic
Sampling temperature
OpinionsQA, Althing persona. Band: ±1 standard deviation across repeat runs.
Persona prompt template
SubPOP, Althing persona, temperature 0.85.
Holdout check
Gap between scores on public questions and on the private answer key. Gaps above 0.05 are flagged for review. The threshold is wider when only a few questions are private.
Runs
The published runs behind these numbers.