This paper considers the dynamic admission control policy in a two-class loss system. Each class of jobs requires
different service rates and offer different rewards. The service provider cannot directly determine the identities but can
observe the signals of the jobs in a batch. The submodularity and concavity properties of the value function are proved.
There is a signal threshold such that the jobs with signals larger than or equal to it are classified as class 1, and those with
signals smaller than it are classified as class 2. Consequently, a four-layer admission control policy is established. When the
signals are less informative, the main results are also available under some certain conditions. Finally, the resulting admission
control policy is applied to an inventory rationing problem with imperfect information, and the feasibility and effectiveness
of such a polity is identified.