Quick Pick Survey asks people to choose between two items and ranks the list from the answers. We have three ways of choosing the next pair. On 2026-10-27 we swept the share of random answers from 5% to 50% on a 30-item list.
The hybrid selector (the one shipped) stayed within 0.005 rank accuracy of the old selector everywhere, and led it by 0.018 at 300 answers when half the answers were random. The pure information-gain selector got worse than the old one once a third of answers were random, because it keeps re-asking pairs whose answers contradict each other.
This is synthetic: random slips, no biased respondents. Table and method are in ventures/quick-pick-survey/docs/INFORMATION-GAIN-SELECTOR.md.