How much error is unavoidable when three yes-or-no variables are modelled by a mixture of just two independent patterns? This anonymous, unrefereed candidate gives an exact answer: about zero point eight three one bits of forward Kullback Leibler divergence in the worst case. Only the two uniform parity distributions reach that limit. Parity means that the number of ones is always even, or always odd. The proposed constant was already recorded in a review by Guido Montufar; this release offers a computer-assisted proof candidate, not a claim to have invented the constant. The proof first reverses simple local random changes to reduce every target to six kinds of support. It then covers the remaining continuous spaces with sixteen thousand six hundred pieces. Each piece has one explicit two-component mixture that works throughout it. Exact arithmetic checks the vertices, while convexity carries their bounds to every point inside. A short algebraic argument supplies the matching parity lower bound. The bundle also turns this construction into a program: supply rational probabilities and receive exact feasible mixture parameters and a trace. Conditioning transfers the bound to a specified family of larger product-mixture models, without proving their worst-case error is exactly the same. This is the Evidence Press release of fifth September twenty twenty-six. The paper, certificates, source, review response and assurance limits are linked on this page. Producer checks and internal model review are not independent mathematical validation. This OpenAI synthetic voice is a communication aid, not additional evidence.