Successful Decode Is Not Referent Resolution: Session Law for Agreement That Is Not Independent Validation

This design paper specifies laboratory law for a decode that succeeds and is still the wrong referent. A sister deposit (10.5281/zenodo.22857303) covers a decoder that remains present and refuses the encode. That refusal rule does not fire when reconstruction produces a usable scene. Successful decoding is not correct referent resolution. Agreement of two sandboxed workers is not independent validation when those workers share a prior. An overloaded token can yield a coherent scene that is a high-frequency mapping rather than the referent the encode issued. Encode-side actuation on that scene is stopped until the issued referent is checked. The mapping is not written into the protected core. Neither decoder is retrained. Three cases are distinguished from walk-away and staying refusal: S4 underspecified token completed from a prior; S5 shared-prior agreement; S6 disagreement at one issued referent. The laboratory demonstration is the token Gilgamesh. One legitimate mapping is the king of Uruk. The issued referent in the session that exposed the hole was Sterling and Zima’s 2002 NASA processor-in-memory architecture, NTRS 20060029890. Two language-model workers, in separate decoding sessions, each selected the king. Agreement did not validate the mapping. Falsifiers F-A1 through F-A5 are stated for a bounded room. Satisfaction of the sister deposit’s refusal falsifiers does not close this hole. This manuscript is not a new QNN architecture and does not license a CAVE installation.