A fast structural gate for RAG that reduces wrong-entity and “Frankenstein context” failures. This paper introduces TL Filter, a drop-in middleware layer derived from Triad Logic (TL) that verifies retrieved context before generation. TL Filter operationalizes three closure conditions—Identity (stable referents), Distinction (non-collapse / optional typed disjointness diagnostics), and Binding (query-connectedness)—by building an induced constraint graph over mentions and chunks. The system filters disconnected snippets, provides explainable drop reasons, and optionally proposes bridge search (bounded multi-hop retrieval) to reconnect relevant information. The paper includes a paper-executable algorithm specification, a production-shaped Python reference implementation, integration guidance, and an evaluation template with metrics targeting entity drift and unsupported narrative stitching.