Nine Star Trek questions from a Rebase round, put to TypeSafe's Jev as text. It answered five correctly. On the other four it returned nothing, so it never gave a wrong answer.
The ledger
Each pill is filled to the probability Jev gave it, on one scale across all nine rounds. Jev gets an unsure option in every round, drawn as the fifth pill. Alongside the probabilities it returns a separate confidence figure, which measures how concentrated the distribution is. A round is answered on that confidence, not on the winning probability, so a leading option can sit above 0.40 while the round still returns nothing.