We revisit the problem of \textit{online linear optimization} in case the set
of feasible actions is accessible through an approximated linear optimization
oracle with a factor $\alpha$ multiplicative approximation guarantee. This
setting is in particular interesting since it captures