# Quantum v4.23 — independent stages and post-analyzer coupling Local scoped/gamma candidate on `2f56a1a3d80a95a01277fcfaa2693e477c4f5f41` (v4.22), branch `quantum-postmeasurement-v423`. Published Leanbook `v1.0.0` remains unchanged. Toolchain and pinned Mathlib are unchanged. [Exact inventory](quantum-v4.23-inventory.json) · [Verification record](quantum-v4.23-verification.json) ## Independent environment stages `QuantumTensorStageEnvironment` defines `tensorGate`, `tensorStage`, `stageStep` and `tensorStages`. A list of pairs `Stage n × Stage e` allows each step to have its own environment unitary, CPTP channel and finite Kraus-label type. The theorem `traceEnvironment_run_tensorStages` eliminates this independent environment activity under partial trace for every joint matrix, without a PSD or product premise. `tensorStages_outcome_marginal_reduction` lifts the result to all five outcomes and arbitrary readout. The uniform-environment corollary matches the proposed `ss.map` interface. `tensorStage_identity_step_eq_extendStage` and `tensorStages_identity_run` compare actions with v4.22. They do not equate `labels × Unit` and `labels` as Stage data. Regression fixtures include heterogeneous label types, arbitrary environment unitaries and nonunital `reset 0`, whose image of identity has entry 2. ## Interaction that the instrument cannot distinguish `QuantumPostBSMEnvironment.postBSMCnotBE` is a CNOT from B to E. It is appended immediately before computational BC readout, after the preceding chronological run. `postBSMCnotBE_not_kronecker_product` excludes every factorization `A ⊗ B`, even without requiring either proposed factor to be unitary. The cancellation lemma uses `computationalReduction m`, or its corrected operator `correction r * computationalReduction m`. It deliberately does not use `branchOperator`, which already includes the Bell analyzer. Within a fixed computational branch both sides have the same measured B value. The environment index is permuted equally on both sides; AD coherences remain intact. `post_interaction_outcome_invariant` proves equality of all five reduced outcomes for arbitrary Readout, arbitrary joint matrix, and **any preceding joint stage list**. This does not make arbitrary preceding interactions marginally reducible. The specialized `postBSMCnotBE_preserves_all_reducedJointOutcomes` derives marginal dependence for an extended system-only Bell analyzer followed by this nonfactorizing interaction. A normalized PSD pure probe is specified at this measurement cut. Its joint matrix changes under the CNOT. The transformed matrix is the outer product of `coupledKet`, and `coupledKet_not_product` excludes a system/environment vector factorization. This is a pure-state obstruction; no general mixed-state separability predicate is introduced. ## Boundaries and regression history The two modules contain 30 named theorem declarations (9 and 21), including helpers and two registry suites. The inventory exports 45 named source declarations including definitions and suite structures. These are code metrics, not counts of independent scientific results. The original external-qubit CNOT A-to-E claim is false and is not restored. Its optional separate coherence-counterexample theorem is not part of this candidate. A single changed output on one input would not by itself establish the absence of every marginal-only map. The earlier v4.21/v4.22 witnesses retain their precise two-input obstructions and placement constraints. No full classification of interactions, adaptive process tensor, or hardware claim is made. Independent-environment reduction is sufficient, not necessary. Initial correlations may persist in the full joint state. Validation is scoped/gamma. The exact report records strict elaboration, incremental integration build, imported axiom closure, independent compiled project audit and targeted container tests. No fresh full `verify-all`, full unittest discovery, clean build or remote CI run is claimed. The last full validation checkpoint remains v4.21 with its documented multi-run test history; v4.22 and v4.23 add bounded evidence only.