Relative clauses and tensed clauses are standardly assumed to be scope islands. However, naturally-occurring counterexamples are abundant and easy to find. Therefore we should revisit analyses that rejected Quantifier Raising on the assumption that QR is clause bounded. The data show that scope islands are sensitive to the identity of both the scope-taker and the predicate embedding the island. I propose the Scope Island Subset Constraint: given two scope islands, the scope-takers trapped by one will be a subset of the scope-takers trapped by the other. A simple refinement of semantic types allows encoding and enforcing scope islands.