We don't show a percentile for a test until 300 people have completed it. Below that, any comparison would be a guess dressed up as a statistic — so results, share cards, and the leaderboard all show a beta state instead until the real number is ready.
Each question carries calibration data — how hard it is and how well it distinguishes stronger from weaker performance — refined from real sittings over time. This page will expand to cover that pipeline in detail as it comes online.