Excision is a principle saying that, under the right cover, removing the same subspace from a space and a subspace should not change relative homology. This anonymous, unrefereed candidate gives a four-point failure for one precise cubical theory of closure spaces. Start with the directed interval J plus and form its inductive square. Let A contain the top and right points, and let B contain the bottom and left points. Their interiors cover all four points. The intersection L consists of two discrete points. Relative to L, B has two directed edges from zero zero. Their difference is a nonzero one-cycle, and an integer cocycle proves that it generates a copy of the integers. Relative to A, the same cycle becomes the boundary of the identity square. Thus the inclusion sends the nonzero group Z to the zero group, so the excision map is not injective. The claim is deliberately narrow: it concerns integral normalized cubical homology for the directed interval J plus with the inductive box product. It says nothing about the categorical product, the other interval objects, symmetric digital images, or other coefficient conventions. Python and JavaScript agree on the complete stable certificate, six convention mutations are rejected, and a fresh replay leaves all fifty-six package files unchanged. Those are producer-side checks, not independent validation, formal verification, specialist review, or peer review. Historical priority also remains unresolved. The scholarly creator is Anonymous. This synthetic-voice briefing is a communication aid, not additional mathematical evidence.