# Quantum v4.27 — Werner cascade reduction and fixed points Local scoped/gamma addition on `d54e87cf10c7f50f2caeb66dcd6dd67c0acfccfc` (v4.26), branch `quantum-werner-v427`. Published Leanbook `v1.0.0` is unchanged. No toolchain, dependency or verifier-policy change. [Exact inventory](quantum-v4.27-inventory.json) · [Verification record](quantum-v4.27-verification.json) ## Operator refinement `QuantumWernerCascade.preStages` applies a specified channel before the entire Bell analyzer, with spectator environment. `preRound_map` proves exact equality with the existing `noisyOutcome` on the system marginal for every matrix and outcome. This fixes the channel insertion cut; it is not an arbitrary model of noise within a physical CNOT gate. `wernerRound` chooses central BC isotropic depolarization, perfect readout and full-pair AD isotropic post depolarization. This is the pair-noise model; existing D-only refinements keep their earlier hypotheses and are not generalized here. `wernerCascade` uses a fixed fresh Werner pair and independent fresh second memory, with second-round parameters identical for every delivered first message. `cascade_visibility_matrix` evaluates the actual `QuantumCascadedSwapping` branches and proves equality of the complete accepted output matrix with an isotropic state of visibility `c₂ * b₂ * c₁ * b₁ * w₁ * w₂ * w_fresh`. The inputs are two Werner pairs independent of the first memory. The memories may have different arbitrary finite dimensions. The first memory is discarded. Kraus preparation supplies both the fresh pair and the fresh second memory. Because the continuation is identical for every first message, complex linearity permits aggregation before the second round. No per-message equality is assumed. The accepted aggregate has trace one and is PSD. Its target overlap is therefore already its normalized fidelity. `cascade_fidelity_transition` identifies this matrix quantity with the scalar transition from the first-round visibility. `cascade_fidelity_nonincrease` then imports the proved scalar inequality. No desired output matrix or scalar product is postulated as a premise. ## Scalar consequences `QuantumCascadeReduction` retains the explicit real parameters and [0,1] bounds. It proves bounds on alpha, preservation of [0,1], nonincrease for nonnegative visibility, strict decrease for w>0 and alpha<1, and the full equality condition: `T(w) = w` iff `w = 0` or `alpha = 1`. The same condition characterizes equality of fidelity. There is no distillation threshold in this model. At alpha<1 the only fixed point is w=0 (F=1/4); at alpha=1 all inputs are fixed. Constant-parameter iteration is `alpha^n * w`. A finite schedule with varying parameters gives their finite product and is nonincreasing. Actual convergence is proved separately with `Tendsto`: for constant alpha<1, visibility tends to zero and fidelity tends to 1/4. For nonnegative inputs the approach is from above; alpha=0 can reach 1/4 after a single step. A varying schedule has no general convergence guarantee without further assumptions. The code uses Function iteration, not the proposed Nat.iterate form, and closes the correct namespace. ## Checks and boundaries The two modules contain 28 named theorems and 42 named declarations in the export inventory, including definitions, structures and suites. These counts are code metrics, not independent scientific results. New regressions check alpha=0 and alpha=1, strict loss at positive visibility, finite iteration, the separately stated limit and its direction. Operator fixtures use memories of dimensions 2 and 3: three input pair visibilities of 1/2 give F=11/32 with ideal gates; two additional factors 1/2 in round two give F=35/128. They also check the output matrix itself and trace one. Prior cascade, feedback, zero-branch, post-noise and environment fixtures remain in the targeted set. Validation is scoped/gamma: strict elaboration of both new modules, imported axiom closure, incremental root build, targeted container regressions and independent compiled-project audit. Exact results and attempt history are in the verification record. No new full verify-all, full test discovery, clean build or remote CI claim. This is not a universal monotonicity law for CPTP channels. Arbitrary readout, postselection, message-dependent resources, correlated fresh inputs and general noise need not obey this visibility product. No distillation circuit, retained memory process tensor or hardware calibration is supplied. The finite-iteration and limit theorems concern the scalar recurrence; the new operator instrument refinement covers exactly the existing two-round cascade.