We present a claim-relative evidence/reference framework for hybrid quantum-classical workflow integrity. Observational indistinguishability yields structural blind regions, distinct from finite-batch statistical misses. Within the declared lattice, a trusted same-batch scalar $R_0$ suffices for conclusion integrity, aggregate $M_0$ for aggregate plus conclusion integrity, and item-aligned binding for item identity. In 3,600 label interventions, feature/prediction views realize exact label-path invariance; all 764 geometry-aligned aggregate-blind rows equal their paired-clean responses, giving zero attack-only increment. For statistical response, the geometry-aligned construction detects 343/2,700 conclusion-changing ($τ\to 0^+$) label interventions with the conformal rule and 1,183/2,700 with the uncorrected union; the original frozen same-item geometry yields 11/2,617 and 43/2,617, respectively. The executed conformal clean false-action rates are 0.048--0.059 descriptively; its finite-sample guarantee requires exchangeability, which the overlapping-draw design violates. The cluster-preserving adaptive stress test (Gate A) reduces response versus matched controls in 25--40 of 40 environment/split cells while retaining conclusion changes. A bounded 165-design-cell ideal-statevector and finite-shot-emulation branch directly instantiates semantic, estimated and observed kernel transitions. The fixed equal-weight design estimates neither deployment prevalence nor QPU, provider or deployed-service assurance.
Comments
12 pages, 2 figures, 2 tables. Supplementary material (20 pages) is available as an ancillary file. Submitted to IEEE Transactions on Dependable and Secure Computing (TDSC). Reproducibility artifact: https://doi.org/10.5281/zenodo.22750616
We present a claim-relative evidence/reference framework for hybrid quantum-classical workflow integrity. Observational indistinguishability yields structural blind regions, distinct from finite-batch statistical misses. Within the declared lattice, a trusted same-batch scalar R0 suffices for conclusion integrity, aggregate M0 for aggregate plus conclusion integrity, and item-aligned binding for item identity. In 3,600 label interventions, feature/prediction views realize exact label-path invariance; all 764 geometry-aligned aggregate-blind rows equal their paired-clean responses, giving zero attack-only increment. For statistical response, the geometry-aligned construction detects 343/2,700 conclusion-changing (τ→0+) label interventions with the conformal rule and 1,183/2,700 with the uncorrected union; the original frozen same-item geometry yields 11/2,617 and 43/2,617, respectively. The executed conformal clean false-action rates are 0.048--0.059 descriptively; its finite-sample guarantee requires exchangeability, which the overlapping-draw design violates. The cluster-preserving adaptive stress test (Gate A) reduces response versus matched controls in 25--40 of 40 environment/split cells while retaining conclusion changes. A bounded 165-design-cell ideal-statevector and finite-shot-emulation branch directly instantiates semantic, estimated and observed kernel transitions. The fixed equal-weight design estimates neither deployment prevalence nor QPU, provider or deployed-service assurance.
Willis (2026) argues that a recursively self-improving system may increase its effective quantum capability without acquiring additional hardware, and that the resulting hybrid classical-quantum execution path becomes progressively harder to characterise exhaustively. The proposed response abandons exhaustive reconstruction as an assurance objective and instead requires that specified properties hold before computation is permitted to affect consequential state, with five quantum execution controls layered beneath substrate-independent constitutional invariants. This paper accepts that architecture and contests two of its load-bearing predicates. The compilation equivalence predicate. GovernanceEquivalent(C(a), C(e)) carries the entire compilation-integrity argument and is left undefined. It cannot be read as semantic equivalence: deciding circuit equivalence is QMA-complete (Janzing, Wocjan & Beth, 2005), and remains so for the shallow circuits typical of transpiled output, while semantic sameness is in any case the wrong target given what transpilation exists to do. The predicate must therefore be evaluated relative to a declared set of governance-relevant properties. That set is itself a governance object, and under recursive self-improvement it is reachable through the toolchain. An attack follows in which a compilation change justified purely on performance grounds renders a governance-relevant property unmeasurable over the executed representation, so that the permit gate continues to evaluate correctly over a test that no longer discriminates. Traceability of the executed artefact to the authorised one does not prevent this, because traceability and property preservation come apart precisely when a transformation is faithful in derivation and lossy in representation. The sampling boundary. The authorised shot count is specified as a ceiling. A ceiling bounds resource consumption; it does not bound evidentiary resolution, which varies in the same direction. Admissibility evaluated on a point estimate is therefore unsound under an adversarial or merely cost-optimising choice of sample size, and error mitigation supplies a second route that widens the uncertainty region while appearing to improve result quality. Evaluating admissibility over a confidence region rather than a point estimate is sound, imposes the necessary floor implicitly, and fails closed without requiring a separately specified minimum. An authorised sequential-sampling rule is a second sound form, subject to the condition that the stopping rule be fixed prior to sampling. Composition. The two results are not independent. The repair proposed for the second introduces a parameter that is vulnerable to the first, and the properties governing sampling can themselves be attenuated out of the property set. Sound sampling governance presupposes governed property-set membership. General form. Both attacks share a shape: an invariant is enforceable only to the extent that the instrument deciding it — its property set, tolerance, confidence level, sample size, or measurability assumption — is unreachable from inside the system it constrains. Under recursive self-improvement there is always gradient in those parameters, because they trade against latency, cost and throughput, which are the objectives the recursion optimises. No adversarial disposition is required. The paper closes by applying its own criterion reflexively. A response whose complaint is that declarations are mistaken for proof cannot end with declarations and leave the matter there. A governance requirement is structural when its violation is observable without relying on the cooperation of the governed system, and declarative when it is not; three of the four requirements proposed here are organisational by that test, and one is not. This work continues the author's earlier treatment of trajectory-level governance and the Constitutional Failure Boundary methodology (Zenodo 10.5281/zenodo.18980403) and of admissibility operators and bounded reachability in probabilistic decision systems (Zenodo 10.5281/zenodo.20057043), extended here from classical to hybrid quantum execution. No implementation is described and none should be inferred. Attribution: the counterfactual authorisation test adapted in §3.6 was proposed by R. Muro. The Layer 3 / Layer 1 redefinition formulation in §3.5 and the authorised sequential-sampling rule in §4.4 are J. M. Willis's, offered in public response to an earlier and shorter statement of these arguments.
Connecting AI-generated quantum programs to executable experiments requires more than syntactically valid code: the workflow must also control execution authority and preserve the basis for interpreting its results. This paper develops a conceptual architecture organized around three responsibilities: AI-assisted circuit proposal, deterministic classical validation, and quantum measurement execution. A quantum-job contract binds circuit artifacts, measurement settings, resource allocations, backend information, and provenance to explicit admission conditions. The design separates permission to run an experiment from evidence that its output answers the intended scientific question. An analytical single-qubit variational example demonstrates this distinction through exact energy calculations, finite-shot variance, and budget-constrained measurement allocation. Responsibility allocation, runtime interception, and staged transformation provide the architectural principles, while uncertainty reporting and reference calculations supply the scientific checks. A prospective evaluation protocol covers accuracy, policy violations, reproducibility, end-to-end cost, and researcher effort. The framework is presented as a design proposal supported by mathematical illustration, not as a tested implementation; its operational benefits and applicability to larger quantum workloads remain to be established.
We construct a supported local spacetime adaptation of the observation-boundary contract of VII-E and the finite hybrid-channel contract of VII-G. The two contracts are proved independently within one laboratory on I-B’s prescribed conical Lorentzian product. For the observation boundary, we prove the effect-pullback criterion, its reference-stable form, the exact sharp-measurement coherence criterion, and a qubit separation of trace distance 1/2 between ideal residual targets with identical records. For hybrid evolution, we prove a complete branch representation, history-sensitive observation closure, spectator consistency, and the exact recovery–constant-complement criterion. An informative classical record consequently excludes recovery of every unknown quantum input, even when the declared residual is also available. The additional contribution is an end-to-end finite observation package that keeps these distinctions visible in supplied spacetime. We provide a compactly supported 3+1-dimensional lift of a four-face transverse adjacency exchange, an exact guarded reversible encoding, and a separately constructed qubit instrument selecting strictly ordered slices without changing the conical metric. A two-step specialization has two histories with the same terminal slice but different conditional residuals. After coarse aggregation, the record remains unchanged while two complete residual targets have the exact half-diamond gap 4(1 − 1/√2)/25. An explicit residual audit enlarges the required input observation space from the diagonal algebra to the full qubit algebra. A destructive finite clock, complete-channel approximation bounds, and a causal route for both payload and classical control complete the local Grade S construction. The package may serve as a precise testbed for distinguishing record-level claims from instrument-level claims in geometric feedback models. The channel and recovery principles themselves are established finite-dimensional facts; the claimed contribution is their supported assembly and the explicit certificates and separations. No metric emergence, gravitational dynamics, field instrument, radiative coefficient, or quantum Ward identity is constructed.
This work separates several physical and theoretical burdens that are often grouped together under “the quantum measurement problem” and develops an experimentally answerable test of one of them: outcome-conditioned continuation. The proposal does not treat measurement as one indivisible physical event. It distinguishes the admission of a measurement interaction and apparatus, durable record formation, later availability of that record, agreement among physical carriers, global outcome standing, quantum-probability weighting, and the rule governing what physical or predictive description continues after an outcome. These burdens can remain open or fail independently. The principal experimental proposal is the Quantum Continuation Closure Test (QCCT). QCCT-I asks whether two physically distinct proper-mixture preparation procedures that are represented by the same admitted standard quantum state can produce different later statistics after the same admitted first measurement. The canonical preparation pair consists of equal-weight Z-basis and X-basis proper mixtures that are represented by the same maximally mixed qubit density operator under standard quantum mechanics. A fixed standard quantum instrument cannot depend on which proper-mixture decomposition produced that state. Published generalized state-update rules provide explicit theory-side foils for which the preparation decomposition can instead remain relevant to the postmeasurement continuation. Ordinary quantum physics is given first refusal. Preparation mismatch, improper-mixture substitution, calibration failure, different first-measurement seats, preparation side channels, ordinary non-Lüders instruments, temporal process memory, and independently warranted expansion of the physical reference must be tested at their own declared scopes before a stronger preparation-indistinguishability violation can be inferred. QCCT therefore does not treat deviation from an ideal Lüders prediction, failure of one fitted instrument, or a preparation-dependent sequential signal as immediate evidence for new quantum dynamics. The comparison architecture escalates from fixed-instrument closure to finite-horizon process-tensor closure and only then to stronger bounded residual claims. The experiment is modular. Prospective QCCT-I execution, process-memory characterization, coarse-graining invariance, archival-data analysis, and later distributed partial closure may contribute distinct local results. A result transfers to a later experiment only through an explicit interface preserving the relevant physical carrier, reference, scope, calibration, provenance, and unresolved burdens. The deposit also contains a machine-auditable formal surface. One consolidated Lean file contains two scientifically distinct Quantum Measurement modules: Seam G, concerning Born weighting / quantum probability measure, and QCCT-I, concerning preparation-indistinguishability / outcome-conditioned continuation. Their shared file packaging does not merge their objects, burdens, evidence states, dispositions, or claims. Both generic machine dispositions remain Not Yet Adjudicable. Seam G retains open domain-owned probability-admission burdens. QCCT-I retains open empirical preparation, reference, measurement, side-channel, comparator, and adjudication burdens. The consolidated Quantum Measurement contracts-and-adapters file was appended directly beneath the unchanged Structural Flow Universal Kernel and elaborated cleanly twice during the release audit. Clean elaboration establishes only that the encoded contracts and adapters coexist coherently at their declared formal scopes. It does not derive the Born rule, establish the physical validity of the generalized continuation family, demonstrate preparation equivalence, close the ordinary quantum comparators, establish an anomaly, falsify standard quantum mechanics, or select an interpretation of quantum measurement. The physical answer remains the responsibility of experiment.
Assertions about collider classifiers require observable evidence and a specified deployment reference. We distinguish the classification threshold that selects events from the claim threshold that defines acceptable performance. For fixed nominal-weighted ratio claims on a frozen finite population, we apply established assorter reductions and bounded-mean confidence sequences under independent sampling with replacement and a predeclared weight bound. The anytime-valid error control is per claim, not automatically simultaneous across claims, and is not a new statistical guarantee. In 19,680 archived paired audit-stream comparisons, changing unweighted correctness and its source reference to nominal-weighted accuracy and its own reference changed 536 supported verdicts to unresolved and 226 in the reverse direction. Scores, labels, weights and exact ratios were available before these simulated audits; their stopping times demonstrate no operational acquisition saving. A post hoc analysis of 30 finite-shot quantum-kernel realizations, with paired RAW and diagonal-loading treatments, describes sensitivity to operating rules, ideal versus realized references, and fixed versus recomposed margin groups. Verdict differences are not auditor error rates, and no quantum-specific mechanism or quantum advantage is established. A CMS open-data example supports restricted aggregate statements while event-level accuracy remains unidentified. This substantial revision withdraws the previous physical-interval coverage conclusions following numerical-validation problems; the available reevaluation does not reestablish them. The retained contribution is mathematical exposition and descriptive evidence, without methodological-priority or calibrated physical-inference claims.
603 — AxoDen-Q FoundationsThis paper presents an early AxoDen-Q research note for hybrid quantum-classical AI safety. It proposes three structural constraints: thermodynamic entropy budgets, epistemic influence bounds, and structural independence requirements. The document formulates candidate compositional bounds and identifies proof obligations needed to move from local stage constraints to system-level guarantees. It is a foundational research note, not an implementation guide. 604 — AxoDen-Q Systems & CertificationThis engineering companion interprets the AxoDen-Q safety envelope as runtime and certification architecture. It outlines a five-layer reference stack, monitoring targets, VSR-Q recovery patterns, CI/CD and policy-as-code controls, and a certification workflow sketch for high-assurance systems. Claims of correctness or assurance are design targets contingent on the companion theory and domain-specific validation. 605 — AxoDen-Q Technical AppendixThis technical appendix develops the broader compositional information-theoretic framework for hybrid classical and quantum AI safety. It covers EBDP, EFI, Q-EBDP, Q-EFI, compositional entropy bounds, model-influence constraints, fabrication-detection statistics, variational reconciliation, estimation approaches, certification protocols, runtime monitoring, and representative domain applications. Any case-study material should be read as protocol structure rather than measured deployment performance.