Research questionHow can variance-inflated Thompson Sampling be analyzed for stochastic generalized linear bandits without tractable posterior approximations?Thompson Sampling analyses for these bandits often require posterior variance inflation, while prior guarantees may rely on tractable posterior approximations. Establishing regret bounds without that approximation assumption is therefore difficult.