Variants (arms)

Simulation speed

UCB1 live state

Total rounds0
Best variant (true)—
UCB1 pick this round—
UCB1 conversions0
UCB1 conv. rate—
Baseline conversions0
Baseline conv. rate—
UCB1 cumulative regret0.00
Baseline cumulative regret0.00
Reward advantage0
Each arm has a hidden true conversion rate the algorithm never sees. UCB1 tracks per-arm pulls and average reward, then each round picks the arm with the highest score avg_reward + √(2·ln(N)/ni) — balancing trying the best-so-far arm (exploitation) against under-sampled arms whose confidence bound is still wide (exploration). The grey towers show a naive uniform 50/50-style baseline run on the identical simulated conversion draws for a fair comparison.