Frankl's union-closed sets conjecture asks whether every finite union-closed family has an element appearing in at least half its sets. The conjecture remains open. This anonymous, unrefereed candidate does not prove a new bound. Instead, it audits one entropy programme and identifies exactly where a proposed structural reduction loses its justification. In Liu's first arXiv version, a joint-concavity inference is used to reach component laws with shared support weights. The candidate gives an explicit perturbation that preserves both aggregate constraints but bends the objective upward. The construction works for every positive protocol parameter up to one, including the sufficiently small regime in the printed theorem. This challenges that inference only. It does not show Liu's theorem false, and it does not affect the separate analytic statement that some non-explicit strict improvement exists. Several useful pieces survive. A compact extreme-point argument recovers a binary latent variable. An operator calculation and exact arithmetic certificate establish the required positive-semidefinite sublemma within a conventional proof boundary. Sequential moment reduction then gives each component its own law with at most three atoms. A sharp endpoint theorem controls entropy-zero limits. A directed-rounding computation certifies one one-component face, and the latent mixing weight can be eliminated exactly. The remaining global problem is still large: ten variables in the generic model after elimination, and six only on a genuine two-by-two subface. The public package contains the manuscript, source map, theorem dependency map, exact and directed-rounding checkers, mutation controls, reference receipts and deterministic archive verifier. These are producer-side and environment-diverse checks. They are not an independent implementation, proof-assistant verification, external specialist review or proof of the global inequality. This synthetic-voice briefing is a communication aid, not additional mathematical evidence.