Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

Readout-Aware Qutrit Feature Geometry: A Cross-Cohort Falsification Study

Matthias Jakob

Published Oct 5, 2026• 0 citations
Publisher

Abstract

Abstract We study whether a fixed qutrit feature geometry built from classical time-series derivatives and residual variability supports compact readouts that remain meaningful across heterogeneous cohorts. The eight special-unitary-group SU(3) Gell-Mann coordinates are derived analytically as deterministic functions of the four classical inputs, making the representation's inductive structure explicit. We then separate representation, candidate readout subspace, fitted direction, target detectability, and predictive utility under participant-separated evaluation. Human Activities and Postural Transitions (HAPT) data repeatedly select the gauge-fixed sector {λ₆, λ₈}. On Daphnet freezing-of-gait data, this core reaches pooled area under the receiver operating characteristic curve (AUROC) 0.72572 with participant-cluster jackknife 95% interval [0.63792, 0.81352], close to 0.72872 for the full eight-coordinate qutrit representation (Q3). A prospectively frozen FoG-STAR lower-back test fails its empirical detectability screen: the best pre-specified model has AUROC 0.50977 [0.45615, 0.56339]. A post-hoc event-level audit using archived fold-specific thresholds finds high nominal event sensitivity but 4.06 false alarms per minute on Daphnet and 27.69 on FoG-STAR primary, excluding a practical detector interpretation. Local encoder perturbations do not rescue the failed primary transfer. Same-information classical interaction features match or exceed reconstructed Q3 on several external settings. Thus neither quantum advantage nor SU(3)-specific predictive gain is supported. The contribution is a falsification-oriented, gauge-aware framework for testing whether compact operator readouts survive cohort, sensor, and task shifts.

DOI
10.21203/rs.3.rs-10686792/v1

Related papers

Quantum Versus Classical Machine Learning for Neuropsychiatric Disorders: A Feature-Matched Multi-Cohort Evaluation

Utsav Poudel, Anuj Nepal, Vyas Viswanath +1 more

Resting-state functional connectivity is widely used in machine learning-based classification of neuropsychiatric disorders. Quantum machine learning models have been proposed for functional connectivity data, but few studies have evaluated them under leave-site-out validation or against matched classical models. We compared six quantum model families with classical ones using identical correlation-derived features across four public cohorts (4,308 subjects). Leave-site-out validation was used for the Autism Brain Imaging Data Exchange (ABIDE I), ADHD-200 (attention-deficit/hyperactivity disorder), and REST-meta-MDD (major depressive disorder) cohorts, while the UCLA Consortium for Neuropsychiatric Phenomics (UCLA-CNP) was evaluated using stratified 10-fold cross-validation. The quantum kernel was statistically indistinguishable from a matched classical kernel across all four cohorts, with the average accuracy difference within 1.4 percentage points at 95% confidence. On ABIDE I, a hybrid quantum convolutional network achieved similar accuracy to its classical counterpart (0.645 vs. 0.641), while a trainable quantum kernel exceeded its classical counterpart by 0.064 in area under the curve (AUC). Quantum convolutional networks also achieved higher AUC than their matched classical networks on this cohort and showed smaller train-test gaps on all four cohorts, without an accuracy gain. Overall, most quantum models matched their classical counterparts, but none consistently outperformed them. This study, among the first quantum evaluations of ABIDE I under leave-site-out validation, provides a feature-matched benchmark of quantum and classical models comprising 284 matched comparisons. We further analyse fidelity-based graph construction and account for the observed quantum-classical parity through effective-dimension and eigenspectrum analyses.

arXiv:2610.21344•Oct 15, 2026•0 citations

Bandwidth, Representation Loss, and Temporal Drift in Quantum Kernel Fraud Detection: A Pre-Specified Falsification Study

Mohamed Faisal Sindhi, Hammaz Buksh

Abstract Quantum kernel results are often reported without holding the input representation, training sample, tuning budget and temporal evaluation protocol fixed against a matched classical kernel. This pre-specified, protocol-locked falsification study uses the Elliptic++ transaction tables (203,769 transactions, 49 time steps) under a locked temporal split (training steps 1–24, validation 25–29, test 30–49 in 4 locked windows). The primary metric is pooled Recall@5%. A frozen full-dimensional gradient-boosting model reached 0.381, and every model, including this one, collapsed in the final window. Graph structure and selective reranking were tested first and falsified. Compressing 182 features to 4, 6 and 8 principal components retained 34%, 43% and 51% of the variance and 64%, 50% and 33% of the full-dimensional model’s pooled Recall@5%. In the primary, default-bandwidth analysis, an entangling fidelity quantum kernel was below the strongest matched classical kernel at 4, 6 and 8 qubits (pooled Recall@5% differences −0.078, −0.203 and −0.146). In a secondary, post hoc bandwidth correction, explicit scale tuning raised the entangling kernel’s Recall@5% by 0.070, 0.121 and 0.109; the corrected kernel remained below the strongest matched classical kernel at 6 and 8 qubits and was statistically unresolved at 4 qubits. The selected scale lay at the lower boundary of the tested grid, and the apparent effect of entanglement depended on bandwidth. No repeatable predictive benefit of the quantum kernel over matched classical kernels was observed in the tested setting; the findings do not extend beyond this dataset, protocol, representation, feature-map family, sample size and bandwidth range.

•Oct 13, 2026•0 citations

When Do Quantum Kernels Help? An Interaction-Order Criterion and Null Results Across Three Biomedical Cohorts

Dhruv sachdeva, Rajeev Gupta

Quantum kernel methods are increasingly applied to clinical prediction, but reported advantages rarely come with ablations, confidence intervals or negative controls. We benchmark a hybrid quantum-classical stacking model under nested cross-validation on three public biomedical cohorts: CRyPTIC tuberculosis drug resistance (12,287 isolates, six drugs), Duke DCE-MRI radiogenomics (n = 922, five endpoints) and MIAS mammography (322 images). Against an all-classical ensemble, the quantum branches change AUC by between -0.0024 and +0.0048 on every tuberculosis and Duke endpoint. Of twenty-four paired comparisons none significantly favours the hybrid, and the one nominally significant difference (p = 0.012, the hybrid below logistic regression on mammography) does not survive correction for multiplicity. Two measurements explain the null: fidelity-kernel off-diagonals decay as 0.5037^n (R^2 = 0.99), and a 277-fold increase in simulated entanglement moves AUC by 0.0148. We recast the inductive-bias argument for quantum kernels as a screen that needs no quantum computation: a quantum feature map can help only if the target depends on high-order feature interactions. The screen measured all six tuberculosis drugs as first-order before any quantum model was run. On synthetic controls an entangling map overtakes tuned classical baselines only at maximal interaction order, and its margin grows with sample size, against the common small-data claim. We also document three evaluation failures in our own pipeline, including a feature-label leak (0.918 AUC where the rebuild gives 0.54-0.59) and stacking leakage worth up to 0.09 AUC. All features, predictions and code are released. Data: https://doi.org/10.5281/zenodo.23344459 . Code: https://doi.org/10.5281/zenodo.23344193 .

•Oct 23, 2026•0 citations

Observational Indistinguishability and Integrity Blind Regions in Hybrid Quantum-Classical Workflows

Roberto Fernández-Barrios, Iker Pastor-López, Amaia Pikatza-Huerga +1 more

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.

cs.CRCross-listedarXiv:2609.17150•Sep 16, 2026•0 citations

Qutrit Quantum Instrument: Logical Lock Plus, Sparse Effective Information Pro, and the Unified Evidence Functional Max

Bsmpx

AbstractThis paper establishes the Qutrit Quantum Instrument (QQI), a training-free structure-first inference architecture in which finite revealed evidence determines route authority before a valid qutrit quantum instrument updates state. The architecture combines complete forward–reverse replay, recursive evidence units, simplicial pruning, bidirectional dimension transport, provenance-bound inheritance, atomic ancestry certification, and a canonical evidence functional within one executable core. Route authority is therefore determined from revealed structure before candidate weighting or quantum-instrument readout. The inherited Logical Lock path remains protected when the evidence does not certify displacement, while newly supported structures may enter through the same replay and geometry requirements. The theoretical construction proves qutrit-instrument validity, measurable active choice, unique geometric transport, legacy safety, and level-controlled intersection-union inference. One implementation was evaluated in eighteen retained sealed contracts spanning nonlinear dynamics, Riemannian geometry, relativistic structure recovery, hypersonic flight, financial prediction, operational weather forecasting, chaotic communication, protein-model routing, visual-cortex encoding, turbulence, and additional physical and computational systems. The empirical results establish several distinct evidence classes. Independent Newton–Leipnik replication produced decisive trajectory-wise superiority against a complete five-comparator family. Official sealed RMLR outputs established family-level superiority across the full six-geometry challenge. Anonymous relativistic experiments attained exact scientific-structure recovery. In Qian Xuesen hypersonic boost-glide, QQI recovered the true aerodynamic dynamics in every sealed trajectory and reduced normalized trajectory error by more than eleven orders of magnitude relative to continuous nonlinear least-squares full-physics fitting. External matched contracts surpassed an official pretrained limit-order-book model, the archive-matched operational ECMWF winner, and a per-record post-hoc IMM oracle. Separate blinded and strong-model audits preserved the leading locked tier in CASP16 and NSD, while mechanism audits attained exact finite-factor and holographic-switchback recovery. Taken together, the results establish one executable rule for deciding when inherited structure should remain authoritative, when new structure has earned displacement authority, and when several surviving structures should remain jointly admissible. Validation designThe empirical program used prespecified primary endpoints, frozen comparator suites, independent experimental units, exact one-sided direction tests, endpoint-level intersection-union inference, and Holm control for prespecified secondary families. Each direct comparison fixed the same raw observations, revealed history, prediction horizon, perturbation condition, partition, independent unit, and evaluation code. Full-capability portfolio comparisons preserved each comparator’s pretrained representation or physical model while fixing raw information and outcome timing. Finite comparator families were tested against every applicable constituent and, where defined, against hindsight or post-hoc oracle selectors. Exact-recovery experiments withheld the scientific structure or role map until scoring. Strong-model parity experiments treated identity with a leading locked comparator as a distinct result class. Boundary results were retained separately from confirmatory superiority claims. Principal findings• Independently replicated Newton–Leipnik trajectory superiority. The Newton–Leipnik double-attractor experiment used an original cohort of 24 independent parameter-and-trajectory blocks and an independently generated 24-block replication. The complete five-comparator intersection-union family passed in both cohorts. Relative to the canonical fixed member, mean sealed-horizon error decreased by 99.27% in the original cohort and 97.75% in the independent replication. Relative to validation-selected STLSQ/SINDy, the reductions were 79.44% and 84.10%. Against nearest-analog local-linear continuation, mean error decreased by 99.993% and 99.985%. Against continuous nonlinear least-squares family fitting, QQI won 21/24 trajectories in the original cohort and 22/24 in the independent replication. The exact one-sided probability values were 1.386×10⁻⁴ and 1.794×10⁻⁵; the Holm-adjusted values were 4.157×10⁻⁴ and 5.382×10⁻⁵. The independently reproduced conjunction therefore establishes trajectory-wise superiority under noisy revealed prefixes. • Complete RMLR six-geometry family superiority. Using official sealed RMLR probability outputs, QQI was evaluated against LEM, ALEM, LCM, AIM, BWM, and PEM, as well as prespecified composite selectors. Against the equal six-geometry pool, post-prefix worst-group NLL decreased by 61.93%. Against the per-seed, per-group post-hoc oracle, it decreased by 41.97%, with 12/12 seed-level wins in both comparisons. Against the group-fixed best geometry, full-online NLL decreased by 39.29%. The geometry-specific reductions were 70.53% for LEM, 70.80% for ALEM, 54.61% for LCM, 67.73% for AIM, 36.27% for BWM, and 44.93% for PEM. Every geometry-specific exact direction test passed. The six-geometry intersection-union probability value was 0.003174, establishing superiority at the complete family level rather than on a single favorable geometry. • Exact relativistic recovery and Qian Xuesen hypersonic dynamics. In primitive-observation general relativity, the Einstein-equation template, formula family, variable roles, column permutations, nonzero structure, and sealed responses were withheld. From anonymous primitive geometric scalars, QQI recovered all 24 sealed tensor-component structures and attained exactly zero clean held-out MSE. It won all 24 blocks against each of PySR, PyOperon, Huber regression, RANSAC regression, and STLSQ. Each Holm-adjusted probability value was 2.980×10⁻⁷. The anonymous special-relativity audit withheld relation names, variable roles, formula families, reference expressions, task ordering, and sealed responses. QQI recovered all 8/8 relations exactly. It recorded 7 wins and 1 numerical tie against PySR, won all 8 tasks against PyOperon, and passed the complete five-comparator intersection-union test. The Qian Xuesen hypersonic boost-glide experiment produced the largest numerical separation. In all 24 independent sealed trajectories, QQI selected the true aerodynamic catalog member and matched the per-trajectory post-hoc catalog oracle within 10⁻¹⁰. Mean normalized full-state trajectory MSE was 1.872×10⁻²⁰, compared with 3.695×10⁻⁹ for continuous nonlinear least-squares full-physics fitting. The comparator-to-QQI error ratio was approximately 1.97×10¹¹, exceeding eleven orders of magnitude. QQI won all 24 paired trajectories with exact one-sided p=5.960×10⁻⁸. All five deployable-comparator intersection-union components passed. Every forecast remained physically valid, and every QQI core audit passed. • Strong operational and external comparisons. On 53,248 purged FI-2010 samples, QQI increased macro-F1 by 18.54% over the official pretrained TLOB checkpoint. Multiclass NLL decreased from 3.171627 to 0.927481, and QQI won 11 of 13 independent purged blocks. At a seven-day lead in extreme-temperature forecasting, the archive-matched operational winner was ECMWF IFS HRES. QQI reduced Brier score by 23.22% across 225 city-date cases. The 40 independent city-week units yielded 29 wins, 10 losses, and 1 tie, with exact one-sided p=0.0016889239. Across 60 independent sealed chaotic-communication records, QQI reduced message NLL from 1.1286 for the per-record post-hoc IMM oracle to 1.0229, a 9.36% reduction. It won 57 records and lost 3, with no ties and exact one-sided p=3.127×10⁻¹⁴. Further direct-performance contracts produced a 37.94% reduction against the strongest deployable four-dimensional hyperchaos comparator, a 99.87% reduction against the mixed-thrust Tsien family, and a 61.93% reduction against the Josephson per-block hindsight corner oracle. Their prespecified confirmatory tests all passed. • Top-tier preservation and exact mechanism audits. On the blinded CASP16 support shared by all five locked roles, QQI attained the same mean official domain lDDT, 0.769740, as Yang-Server group 052. All 16 represented independent targets were exact ties. Since Yang-Server was the official top-ranked locked server comparator, the audit places QQI at the same leading tier on this blinded set. In NSD visual-cortex encoding, QQI exactly matched fixed CLIP ridge across all 8 independent participants at mean noise-ceiling-normalized correlation 0.380999. It simultaneously beat fixed DINOv2 fractional ridge and the concatenated-feature PCA-plus-ridge pipeline in all eight participants after Holm correction. The finite-factor audit attained zero torus error on all 32 sealed blocks, equaled the exact finite-factor oracle to machine precision, and rejected every non-oracle comparator with intersection-union p=2.328×10⁻¹⁰. The holographic benchmark selected the counterterm switchback branch exactly in all 32 held-out conditions and tied the catalog oracle. On JHTDB turbulence, QQI reproduced the validation-selected multiresolution predictor at every sealed velocity-field cell and beat four alternative field predictors at all ten sealed origins. Confirmatory scope The five-contract cross-domain intersection-union probability value was 0.01123047. Every component superiority statement in that global claim therefore passed at the five-percent level. The RMLR six-geometry family was tested separately and passed with p=0.003174. In the six-world controlled confirmation, the prespecified all-comparator primary claim passed with intersection-union p=0.003174. Mean NLL and worst-world Brier s

•Aug 17, 2026•0 citations

Auditable Null Results in Quantum Hardware Scaling: A Pre-Specified Falsification Study of QAOA Recovery Models

Devin Peters

Empirical scaling laws for the Quantum Approximate Optimization Algorithm (QAOA) on near-term hardware are often motivated by noise-compounding arguments or circuit-structure heuristics. Such models are rarely evaluated under pre-specified failure criteria that prevent post-hoc rescue. This study reports an audit-disciplined attempt to construct and test a predictive scaling law for QAOA-based minimal cut set recovery in fault tree models. The workflow used hash-locked artifacts, pre-specified kill switches, one-time holdout evaluation, and explicit no-rerun and no-rescue rules at every phase boundary. The initial F²-grounded compounding specification failed its pre-specified identifiability gate (KS3). Under an anti-leakage filter applied to 108 candidate rescue terms, zero candidates recovered a passing α₁ slope while also adding explanatory power above the structural baseline. A pre-specified structure-dominated fallback model was then defined and fit. The fallback reproduced its locked training prediction pathway to machine precision, with maximum absolute prediction difference of 3.55 × 10⁻¹⁵. It then failed the locked 24-row transport holdout. Holdout RMSE was 5.2214 against a locked threshold of 0.3770, and all 24 residuals were positive, giving a two-sided sign-test p-value of 1.19 × 10⁻⁷. The mean log residual of 5.198 corresponds to approximately 180-fold underprediction on the linear scale. Post-failure diagnostics indicated lower-n domain shift with standardized mean difference of −2.27 and four holdout rows with G_2q below the training range. By design, these diagnostics were explanatory only and did not authorize rescue, refit, reweighting, row removal, or threshold revision. The results do not support a deployable scaling law on the present corpus. Instead, they demonstrate that pre-specified audit controls can convert an attempted positive scaling-law claim into a reproducible falsification result with clear claim boundaries. The audit framework, the locked failure chain, and the allowed-and-forbidden claim boundary together constitute the paper's contribution. The work is offered as a methodological reference for future QAOA-on-hardware scaling studies.

•May 13, 2026•0 citations
Abstract We study whether a fixed qutrit feature geometry built from classical time-series derivatives and residual variability supports compact readouts that remain meaningful across heterogeneous cohorts. The eight special-unitary-group SU(3) Gell-Mann coordinates are derived analytically as deterministic functions of the four classical inputs, making the representation's inductive structure explicit. We then separate representation, candidate readout subspace, fitted direction, target detectability, and predictive utility under participant-separated evaluation. Human Activities and Postural Transitions (HAPT) data repeatedly select the gauge-fixed sector {λ₆, λ₈}. On Daphnet freezing-of-gait data, this core reaches pooled area under the receiver operating characteristic curve (AUROC) 0.72572 with participant-cluster jackknife 95% interval [0.63792, 0.81352], close to 0.72872 for the full eight-coordinate qutrit representation (Q3). A prospectively frozen FoG-STAR lower-back test fails its empirical detectability screen: the best pre-specified model has AUROC 0.50977 [0.45615, 0.56339]. A post-hoc event-level audit using archived fold-specific thresholds finds high nominal event sensitivity but 4.06 false alarms per minute on Daphnet and 27.69 on FoG-STAR primary, excluding a practical detector interpretation. Local encoder perturbations do not rescue the failed primary transfer. Same-information classical interaction features match or exceed reconstructed Q3 on several external settings. Thus neither quantum advantage nor SU(3)-specific predictive gain is supported. The contribution is a falsification-oriented, gauge-aware framework for testing whether compact operator readouts survive cohort, sensor, and task shifts.
Resting-state functional connectivity is widely used in machine learning-based classification of neuropsychiatric disorders. Quantum machine learning models have been proposed for functional connectivity data, but few studies have evaluated them under leave-site-out validation or against matched classical models. We compared six quantum model families with classical ones using identical correlation-derived features across four public cohorts (4,308 subjects). Leave-site-out validation was used for the Autism Brain Imaging Data Exchange (ABIDE I), ADHD-200 (attention-deficit/hyperactivity disorder), and REST-meta-MDD (major depressive disorder) cohorts, while the UCLA Consortium for Neuropsychiatric Phenomics (UCLA-CNP) was evaluated using stratified 10-fold cross-validation. The quantum kernel was statistically indistinguishable from a matched classical kernel across all four cohorts, with the average accuracy difference within 1.4 percentage points at 95% confidence. On ABIDE I, a hybrid quantum convolutional network achieved similar accuracy to its classical counterpart (0.645 vs. 0.641), while a trainable quantum kernel exceeded its classical counterpart by 0.064 in area under the curve (AUC). Quantum convolutional networks also achieved higher AUC than their matched classical networks on this cohort and showed smaller train-test gaps on all four cohorts, without an accuracy gain. Overall, most quantum models matched their classical counterparts, but none consistently outperformed them. This study, among the first quantum evaluations of ABIDE I under leave-site-out validation, provides a feature-matched benchmark of quantum and classical models comprising 284 matched comparisons. We further analyse fidelity-based graph construction and account for the observed quantum-classical parity through effective-dimension and eigenspectrum analyses.
Abstract Quantum kernel results are often reported without holding the input representation, training sample, tuning budget and temporal evaluation protocol fixed against a matched classical kernel. This pre-specified, protocol-locked falsification study uses the Elliptic++ transaction tables (203,769 transactions, 49 time steps) under a locked temporal split (training steps 1–24, validation 25–29, test 30–49 in 4 locked windows). The primary metric is pooled Recall@5%. A frozen full-dimensional gradient-boosting model reached 0.381, and every model, including this one, collapsed in the final window. Graph structure and selective reranking were tested first and falsified. Compressing 182 features to 4, 6 and 8 principal components retained 34%, 43% and 51% of the variance and 64%, 50% and 33% of the full-dimensional model’s pooled Recall@5%. In the primary, default-bandwidth analysis, an entangling fidelity quantum kernel was below the strongest matched classical kernel at 4, 6 and 8 qubits (pooled Recall@5% differences −0.078, −0.203 and −0.146). In a secondary, post hoc bandwidth correction, explicit scale tuning raised the entangling kernel’s Recall@5% by 0.070, 0.121 and 0.109; the corrected kernel remained below the strongest matched classical kernel at 6 and 8 qubits and was statistically unresolved at 4 qubits. The selected scale lay at the lower boundary of the tested grid, and the apparent effect of entanglement depended on bandwidth. No repeatable predictive benefit of the quantum kernel over matched classical kernels was observed in the tested setting; the findings do not extend beyond this dataset, protocol, representation, feature-map family, sample size and bandwidth range.
Quantum kernel methods are increasingly applied to clinical prediction, but reported advantages rarely come with ablations, confidence intervals or negative controls. We benchmark a hybrid quantum-classical stacking model under nested cross-validation on three public biomedical cohorts: CRyPTIC tuberculosis drug resistance (12,287 isolates, six drugs), Duke DCE-MRI radiogenomics (n = 922, five endpoints) and MIAS mammography (322 images). Against an all-classical ensemble, the quantum branches change AUC by between -0.0024 and +0.0048 on every tuberculosis and Duke endpoint. Of twenty-four paired comparisons none significantly favours the hybrid, and the one nominally significant difference (p = 0.012, the hybrid below logistic regression on mammography) does not survive correction for multiplicity. Two measurements explain the null: fidelity-kernel off-diagonals decay as 0.5037^n (R^2 = 0.99), and a 277-fold increase in simulated entanglement moves AUC by 0.0148. We recast the inductive-bias argument for quantum kernels as a screen that needs no quantum computation: a quantum feature map can help only if the target depends on high-order feature interactions. The screen measured all six tuberculosis drugs as first-order before any quantum model was run. On synthetic controls an entangling map overtakes tuned classical baselines only at maximal interaction order, and its margin grows with sample size, against the common small-data claim. We also document three evaluation failures in our own pipeline, including a feature-label leak (0.918 AUC where the rebuild gives 0.54-0.59) and stacking leakage worth up to 0.09 AUC. All features, predictions and code are released. Data: https://doi.org/10.5281/zenodo.23344459 . Code: https://doi.org/10.5281/zenodo.23344193 .
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 R0R_0R0​ suffices for conclusion integrity, aggregate M0M_0M0​ 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+τ\to 0^+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.
AbstractThis paper establishes the Qutrit Quantum Instrument (QQI), a training-free structure-first inference architecture in which finite revealed evidence determines route authority before a valid qutrit quantum instrument updates state. The architecture combines complete forward–reverse replay, recursive evidence units, simplicial pruning, bidirectional dimension transport, provenance-bound inheritance, atomic ancestry certification, and a canonical evidence functional within one executable core. Route authority is therefore determined from revealed structure before candidate weighting or quantum-instrument readout. The inherited Logical Lock path remains protected when the evidence does not certify displacement, while newly supported structures may enter through the same replay and geometry requirements. The theoretical construction proves qutrit-instrument validity, measurable active choice, unique geometric transport, legacy safety, and level-controlled intersection-union inference. One implementation was evaluated in eighteen retained sealed contracts spanning nonlinear dynamics, Riemannian geometry, relativistic structure recovery, hypersonic flight, financial prediction, operational weather forecasting, chaotic communication, protein-model routing, visual-cortex encoding, turbulence, and additional physical and computational systems. The empirical results establish several distinct evidence classes. Independent Newton–Leipnik replication produced decisive trajectory-wise superiority against a complete five-comparator family. Official sealed RMLR outputs established family-level superiority across the full six-geometry challenge. Anonymous relativistic experiments attained exact scientific-structure recovery. In Qian Xuesen hypersonic boost-glide, QQI recovered the true aerodynamic dynamics in every sealed trajectory and reduced normalized trajectory error by more than eleven orders of magnitude relative to continuous nonlinear least-squares full-physics fitting. External matched contracts surpassed an official pretrained limit-order-book model, the archive-matched operational ECMWF winner, and a per-record post-hoc IMM oracle. Separate blinded and strong-model audits preserved the leading locked tier in CASP16 and NSD, while mechanism audits attained exact finite-factor and holographic-switchback recovery. Taken together, the results establish one executable rule for deciding when inherited structure should remain authoritative, when new structure has earned displacement authority, and when several surviving structures should remain jointly admissible. Validation designThe empirical program used prespecified primary endpoints, frozen comparator suites, independent experimental units, exact one-sided direction tests, endpoint-level intersection-union inference, and Holm control for prespecified secondary families. Each direct comparison fixed the same raw observations, revealed history, prediction horizon, perturbation condition, partition, independent unit, and evaluation code. Full-capability portfolio comparisons preserved each comparator’s pretrained representation or physical model while fixing raw information and outcome timing. Finite comparator families were tested against every applicable constituent and, where defined, against hindsight or post-hoc oracle selectors. Exact-recovery experiments withheld the scientific structure or role map until scoring. Strong-model parity experiments treated identity with a leading locked comparator as a distinct result class. Boundary results were retained separately from confirmatory superiority claims. Principal findings• Independently replicated Newton–Leipnik trajectory superiority. The Newton–Leipnik double-attractor experiment used an original cohort of 24 independent parameter-and-trajectory blocks and an independently generated 24-block replication. The complete five-comparator intersection-union family passed in both cohorts. Relative to the canonical fixed member, mean sealed-horizon error decreased by 99.27% in the original cohort and 97.75% in the independent replication. Relative to validation-selected STLSQ/SINDy, the reductions were 79.44% and 84.10%. Against nearest-analog local-linear continuation, mean error decreased by 99.993% and 99.985%. Against continuous nonlinear least-squares family fitting, QQI won 21/24 trajectories in the original cohort and 22/24 in the independent replication. The exact one-sided probability values were 1.386×10⁻⁴ and 1.794×10⁻⁵; the Holm-adjusted values were 4.157×10⁻⁴ and 5.382×10⁻⁵. The independently reproduced conjunction therefore establishes trajectory-wise superiority under noisy revealed prefixes. • Complete RMLR six-geometry family superiority. Using official sealed RMLR probability outputs, QQI was evaluated against LEM, ALEM, LCM, AIM, BWM, and PEM, as well as prespecified composite selectors. Against the equal six-geometry pool, post-prefix worst-group NLL decreased by 61.93%. Against the per-seed, per-group post-hoc oracle, it decreased by 41.97%, with 12/12 seed-level wins in both comparisons. Against the group-fixed best geometry, full-online NLL decreased by 39.29%. The geometry-specific reductions were 70.53% for LEM, 70.80% for ALEM, 54.61% for LCM, 67.73% for AIM, 36.27% for BWM, and 44.93% for PEM. Every geometry-specific exact direction test passed. The six-geometry intersection-union probability value was 0.003174, establishing superiority at the complete family level rather than on a single favorable geometry. • Exact relativistic recovery and Qian Xuesen hypersonic dynamics. In primitive-observation general relativity, the Einstein-equation template, formula family, variable roles, column permutations, nonzero structure, and sealed responses were withheld. From anonymous primitive geometric scalars, QQI recovered all 24 sealed tensor-component structures and attained exactly zero clean held-out MSE. It won all 24 blocks against each of PySR, PyOperon, Huber regression, RANSAC regression, and STLSQ. Each Holm-adjusted probability value was 2.980×10⁻⁷. The anonymous special-relativity audit withheld relation names, variable roles, formula families, reference expressions, task ordering, and sealed responses. QQI recovered all 8/8 relations exactly. It recorded 7 wins and 1 numerical tie against PySR, won all 8 tasks against PyOperon, and passed the complete five-comparator intersection-union test. The Qian Xuesen hypersonic boost-glide experiment produced the largest numerical separation. In all 24 independent sealed trajectories, QQI selected the true aerodynamic catalog member and matched the per-trajectory post-hoc catalog oracle within 10⁻¹⁰. Mean normalized full-state trajectory MSE was 1.872×10⁻²⁰, compared with 3.695×10⁻⁹ for continuous nonlinear least-squares full-physics fitting. The comparator-to-QQI error ratio was approximately 1.97×10¹¹, exceeding eleven orders of magnitude. QQI won all 24 paired trajectories with exact one-sided p=5.960×10⁻⁸. All five deployable-comparator intersection-union components passed. Every forecast remained physically valid, and every QQI core audit passed. • Strong operational and external comparisons. On 53,248 purged FI-2010 samples, QQI increased macro-F1 by 18.54% over the official pretrained TLOB checkpoint. Multiclass NLL decreased from 3.171627 to 0.927481, and QQI won 11 of 13 independent purged blocks. At a seven-day lead in extreme-temperature forecasting, the archive-matched operational winner was ECMWF IFS HRES. QQI reduced Brier score by 23.22% across 225 city-date cases. The 40 independent city-week units yielded 29 wins, 10 losses, and 1 tie, with exact one-sided p=0.0016889239. Across 60 independent sealed chaotic-communication records, QQI reduced message NLL from 1.1286 for the per-record post-hoc IMM oracle to 1.0229, a 9.36% reduction. It won 57 records and lost 3, with no ties and exact one-sided p=3.127×10⁻¹⁴. Further direct-performance contracts produced a 37.94% reduction against the strongest deployable four-dimensional hyperchaos comparator, a 99.87% reduction against the mixed-thrust Tsien family, and a 61.93% reduction against the Josephson per-block hindsight corner oracle. Their prespecified confirmatory tests all passed. • Top-tier preservation and exact mechanism audits. On the blinded CASP16 support shared by all five locked roles, QQI attained the same mean official domain lDDT, 0.769740, as Yang-Server group 052. All 16 represented independent targets were exact ties. Since Yang-Server was the official top-ranked locked server comparator, the audit places QQI at the same leading tier on this blinded set. In NSD visual-cortex encoding, QQI exactly matched fixed CLIP ridge across all 8 independent participants at mean noise-ceiling-normalized correlation 0.380999. It simultaneously beat fixed DINOv2 fractional ridge and the concatenated-feature PCA-plus-ridge pipeline in all eight participants after Holm correction. The finite-factor audit attained zero torus error on all 32 sealed blocks, equaled the exact finite-factor oracle to machine precision, and rejected every non-oracle comparator with intersection-union p=2.328×10⁻¹⁰. The holographic benchmark selected the counterterm switchback branch exactly in all 32 held-out conditions and tied the catalog oracle. On JHTDB turbulence, QQI reproduced the validation-selected multiresolution predictor at every sealed velocity-field cell and beat four alternative field predictors at all ten sealed origins. Confirmatory scope The five-contract cross-domain intersection-union probability value was 0.01123047. Every component superiority statement in that global claim therefore passed at the five-percent level. The RMLR six-geometry family was tested separately and passed with p=0.003174. In the six-world controlled confirmation, the prespecified all-comparator primary claim passed with intersection-union p=0.003174. Mean NLL and worst-world Brier s
Empirical scaling laws for the Quantum Approximate Optimization Algorithm (QAOA) on near-term hardware are often motivated by noise-compounding arguments or circuit-structure heuristics. Such models are rarely evaluated under pre-specified failure criteria that prevent post-hoc rescue. This study reports an audit-disciplined attempt to construct and test a predictive scaling law for QAOA-based minimal cut set recovery in fault tree models. The workflow used hash-locked artifacts, pre-specified kill switches, one-time holdout evaluation, and explicit no-rerun and no-rescue rules at every phase boundary. The initial F²-grounded compounding specification failed its pre-specified identifiability gate (KS3). Under an anti-leakage filter applied to 108 candidate rescue terms, zero candidates recovered a passing α₁ slope while also adding explanatory power above the structural baseline. A pre-specified structure-dominated fallback model was then defined and fit. The fallback reproduced its locked training prediction pathway to machine precision, with maximum absolute prediction difference of 3.55 × 10⁻¹⁵. It then failed the locked 24-row transport holdout. Holdout RMSE was 5.2214 against a locked threshold of 0.3770, and all 24 residuals were positive, giving a two-sided sign-test p-value of 1.19 × 10⁻⁷. The mean log residual of 5.198 corresponds to approximately 180-fold underprediction on the linear scale. Post-failure diagnostics indicated lower-n domain shift with standardized mean difference of −2.27 and four holdout rows with G_2q below the training range. By design, these diagnostics were explanatory only and did not authorize rescue, refit, reweighting, row removal, or threshold revision. The results do not support a deployable scaling law on the present corpus. Instead, they demonstrate that pre-specified audit controls can convert an attempted positive scaling-law claim into a reproducible falsification result with clear claim boundaries. The audit framework, the locked failure chain, and the allowed-and-forbidden claim boundary together constitute the paper's contribution. The work is offered as a methodological reference for future QAOA-on-hardware scaling studies.