Testing satisfaction of guards is the essential operation of concurrent constraint programming (CCP) systems. We present and prove correct, for the first time, an incremental algorithm for the simultaneous tests of entailment and disentailment of rational tree constraints to be used in CCP systems with deep guards (e.g., AKL or Oz). The algorithm is presented as the simplification of the constraints which form the (possibly deep) guards and which are situated at different nodes (or, local computation spaces) in a tree (of arbitrary depth). In this algorithm, each variable may have multiple bindings (representing multiple constraints on the same variable in different nodes). These may be realized by re- and de-installation upon each newly resumed check of the guard in the corresponding node (as done, e.g., in AKL or Oz), or by using look-up tables (with entries indexed by the nodes). We give a simple fixed-point algorithm and use it for proving that the tests implemented by another, practical algorithm are correct and complete for entailment and disentailment. We formulate the results in this paper for rational tree constraints; they can be adapted to finite and feature trees.