# Quantum v4.28: adaptive Werner cascade with detector loss Local additive candidate over `3cdac4021bf5c9b574116c19dec2461125b9e31c`. The published Leanbook 1.0.0 tag is unchanged. The first result proves each individual corrected Bell sandwich, then includes central BC depolarization, full-pair AD post depolarization, and faithful delivery with detector efficiencies in [0,1]. It does not infer individual branches from an aggregated equality. Each initial Werner pair is independent of the other and of the first memory. The actual v4.26 Cascade chooses second-round noise, a fresh Werner pair and fresh independent memory by the delivered first message. Each successful history has weight eta1(r)*eta2(r,t)/16. Summing gives q(r) and total success P. The exact matrix sum, trace P, zero matrix at P=0 and normalized affine Werner matrix at P>0 are proved. Conditional visibility is bounded in [0,1]. Selected branches also have an explicit conditional matrix formula. The target-overlap lemma connects these matrices to the scalar fidelity comparison. A positively weighted branch above the weighted mean beats the aggregate fidelity but cannot exceed the first-round fidelity. This is not global policy optimization or entanglement distillation. Arbitrary label confusion is outside the Werner specialization; the general underlying instrument retains that capability. The affine `wernerMatrix` accepts real scalars algebraically; physical operator channels use bounded `Visibility`. The conditional bound is separately proved. First memory is discarded; no persistent multi-round environmental correlations or process tensor model are claimed. Validation scope is gamma: strict module/import closure, incremental registry build and targeted regressions. See the verification JSON for actual results; these do not constitute a new full-library test discovery or clean build.