Bounded Rewards

Assumption - Rewards are uniformly bounded.

Applies to - IS and related estimators when deriving concentration bounds.

Definition - There exists R_max such that |R_t| <= R_max almost surely.

Diagnostics - Check empirical reward range and clip if necessary.

Failure modes - Heavy-tailed rewards inflate variance and invalidate CI assumptions.

References - Uehara, Shi, Kallus (2022).