The mathematical heart of the framework is the Fiedler eigenvalue, denoted lambda-2, the second-smallest eigenvalue of the normalized hypergraph Laplacian. In spectral graph theory, this quantity measures a network’s algebraic connectivity: how readily information, or in this case procurement transactions and feature samples, can mix across the whole structure. The authors prove a theorem establishing that as lambda-2 approaches zero, the relaxation time of a random walk on the network, given by two divided by lambda-2, diverges. Because SHAP-style estimators draw background samples from the empirical distribution of supply-network transactions, a collapsing network traps those samples in isolated connected components. The result is a proof that lambda-2 tending to zero is a sufficient condition for permutation-based XAI failure: once connectivity collapses, no increase in sampling effort can recover valid attributions, because the estimator becomes inconsistent, with an irreducible error floor.

Crucially, the theorem is a sufficient condition, not a necessary one. Attribution methods can fail for other reasons, such as distribution shift or model misspecification, even when connectivity remains high. Conversely, certifying that lambda-2 exceeds a calibrated threshold only rules out connectivity-induced failure. The authors are explicit that their validity certificate therefore aggregates four spectral governance metrics rather than relying on the eigenvalue alone: a spectral gap metric, an explanation drift index, a topological stability index computed through persistent homology, and the fraction of hyperedges with negative Ollivier-Ricci curvature, a discrete geometric measure of local connectivity.

The choice of a hypergraph rather than a pairwise graph reflects the structure of pharmaceutical regulation. A supply contract in the South African chronic medicines programme binds at minimum a manufacturer, a cold-chain logistics operator and a dispensing cluster under a single regulatory obligation, forming a hyperedge of three or more parties. Projecting such contracts onto pairwise edges destroys the joint accountability structure and creates spurious independence assumptions, which the authors argue degrades both connectivity estimates and attribution quality. The framework is deliberately interdisciplinary, drawing on spectral hypergraph theory, the axiomatic foundations of additive attribution methods, and information processing theory, which predicts that explanations become unreliable when the network’s information propagation capacity falls below the demands of global feature attribution.

Empirically, the study is a calibrated simulation, and the authors are unusually emphatic about this limitation. The supply hypergraph itself is entirely synthetic, generated with parameters calibrated to the Central Chronic Medicines Dispensing and Distribution programme. What is real are the shock inputs: a cleaned panel of 3,227 business days spanning 2014 to 2026, combining Eskom daily load-shedding stage records and South African Reserve Bank ZAR/USD exchange rates. The composite shock index built from these two series correlates with the synthetic spectral gap at r = 0.975, but the authors caution that this near-unity correlation is a calibration consistency check, not an independent empirical discovery, since the shock index directly drives the simulated weight reductions. The substantive contribution at this level is that documented South African shock magnitudes are large enough to push the synthetic network into degraded governance regimes.

The framework’s constructive component is lambda-SHAP, a spectral shrinkage estimator that blends current SHAP attributions toward a validated fluid-regime prior, using the time-varying Fiedler eigenvalue as the shrinkage weight. Under normal connectivity the estimator reduces to standard SHAP; as lambda-2 falls, weight shifts toward historically validated attributions. This trades variance for a controlled, documented bias, and the authors flag the directional risk openly: in a genuine structural break where current feature importances differ qualitatively from the past, the estimator may suppress exactly the signals that matter, a false-safety risk mitigated by mandatory human review in the red regime and a prior recalibration protocol.

In an oracle experiment using a linear model with analytically known true Shapley values, lambda-SHAP reduced attribution mean squared error by 70.4 percent in the green regime to 89.2 percent in the red regime, with the advantage growing monotonically as connectivity declined. An ablation study identified the spectral gate as the decisive component: removing it collapsed performance to standard SHAP levels, while the probabilistic certificate layer contributed governance features rather than error reduction. On the real proxy panel, using exchange-rate volatility as an attribution surrogate, lambda-SHAP cut explanation drift index exceedances of the 0.30 regulatory threshold from 9.2 percent to 4.6 percent of red-regime days, and produced zero threshold exceedances during the COVID-19 episode compared with twelve for standard SHAP.

Detection performance also favoured the spectral approach. In Monte Carlo experiments with 500 replications, the topological stability index detected disruptions in a mean of 6.3 business days, versus 28.3 days for GNNExplainer and 38.4 days for a CUSUM change-point detector, with all comparisons statistically significant. The authors attribute part of this advantage to persistent homology detecting path loss before full connectivity loss registers in the spectral gap. An early-warning analysis of the synthetic series found the calibrated amber threshold alert preceded the red-threshold crossing by 19 business days, lead time the authors argue could allow patient re-routing before supply exhaustion, though they stress that quantifying the operational impact requires field data. Extreme value analysis added a further warning: once the network enters the arrested state, the tail of the spectral risk index follows a Generalized Pareto distribution with positive shape parameter, meaning disruption severity becomes structurally unbounded precisely when conventional bounded-tail risk models assume the opposite.

The governance output is a probabilistic validity certificate: a continuous score between zero and one, produced by ridge-regularized logistic regression over the four spectral metrics, with 95 percent bootstrap confidence intervals and SHA-256 tamper-evidence, readable by a procurement officer as a confidence percentage alongside a regime flag and a prescribed action. Within the synthetic setting the certificate was well calibrated, with a cross-validated log-loss of 0.0076, though the authors note that a near-perfect AUC of 1.0 reflects the deterministic coupling built into the simulation and will not transfer to noisy real data. A nine-step, costed deployment pathway, totalling R915,000, describes integration with the existing DHIS2 national health information infrastructure across South Africa’s nine provinces, with optional blockchain anchoring of certificate hashes on a permissioned ledger proposed for later phases.

The limitations are structural and clearly stated. No real pharmaceutical hypergraph was analysed, no field deployment has occurred, and disruptions were injected as parametric weight reductions that may not capture the correlated, non-monotone character of real crises. Field validation on actual procurement transaction data from the chronic medicines programme is planned for 2027 under the PLMIF-2 study, which will test whether the predicted regime-dependent attribution instability appears in a deployed model and whether the proxy-panel reduction estimates hold on real data. If it does, the authors argue the implication extends well beyond medicines: any AI-governed networked system with multi-party, hypergraph-like contracts, from energy dispatch to emergency logistics, faces the same geometric vulnerability, and the reliability of its explanations is bounded above by the algebraic connectivity of the network on which it operates.

Subject of Research: Spectral XAI validity certification for AI-governed pharmaceutical supply chains using the Fiedler eigenvalue

Article Title: SpectralGuard: a Spectrally grounded λ-SHAP framework for XAI validity certification in pharmaceutical supply chain networks

Article References: Moroke, N. D., & Lebele, M. M. (2026). SpectralGuard: a Spectrally grounded λ-SHAP framework for XAI validity certification in pharmaceutical supply chain networks. Frontiers in Artificial Intelligence, 9, Article 1884880. https://doi.org/10.3389/frai.2026.1884880

Image Credits: AI Generated

DOI: 10.3389/frai.2026.1884880

Keywords: explainable AI, SHAP, Fiedler eigenvalue, hypergraph Laplacian, pharmaceutical supply chain, spectral governance, South Africa, load-shedding, lambda-SHAP, validity certificate, AI regulation, network connectivity

The concept of metabolic arrest is borrowed from biology, where it describes how organisms suppress cellular activity when energy becomes scarce. The financial analogue is grounded in a genuine physical principle: Landauer’s principle, formulated in 1961, establishes that every irreversible computational operation dissipates a minimum amount of heat, proportional to the product of Boltzmann’s constant and the system’s absolute temperature. Every trade executed on an exchange—every price signal, settlement record, and position reconciliation—is a physical computation running on silicon that requires electrical power. When the grid fails, the market’s information-processing throughput contracts. The study is careful, however, to frame Landauer’s principle as a heuristic motivation for a specific functional form rather than a literal physical derivation; no claim is made that a financial trade constitutes a strict thermodynamic erasure event, or that a meaningful temperature can be assigned to a financial system.

TENSORnet—which stands for Temporal Entropy-aware Network for Systemic Onset Recognition, and which, despite its name, contains no neural network at all—fuses two components. The first is a VIX-calibrated inhibitory gate: a sigmoid function, fitted by nonlinear least squares using only training-period data, that suppresses the effective information flux in the entropy measure as a function of infrastructure stress. The second is a cross-asset Shannon entropy layer, computed over a 20-day rolling window across seven asset classes on the Johannesburg Stock Exchange: equities, government bonds, gold, platinum, a commodity index, the VIX itself, and Brent crude. In a healthy market, return activity is distributed across assets and entropy is high. Under stress, activity concentrates, entropy collapses, and the gated signal rises—providing a continuous detection variable that the author maps onto three operational regimes, from passive surveillance to full intervention.

The empirical design is unusually disciplined for this literature. The dataset spans 2,838 JSE trading days from January 2015 to April 2026, paired with the official Eskom load-shedding stage record. Because Stage 4 load-shedding was implemented for the first time in South African history only in February 2019, and Stage 6 only in December 2019, the training window (2015–2018) contains no fragile-regime days at all. The study therefore adopts a three-way split: training for gate calibration, validation in 2019 for threshold selection, and a held-out test period from 2020 to 2026 comprising 1,581 days that was touched exactly once. On the primary Stage 3+ stress definition, TENSORnet achieves an F1 score of 99.55 percent and an area under the ROC curve of 0.9996, with just two false positives out of 1,140 resilient test days.

But the study’s most consequential finding is a negative one. A formal DeLong test comparing TENSORnet’s ROC curve against a naive classifier that simply uses the lagged infrastructure stage finds no statistically significant difference between them. In other words, the elaborate thermodynamically motivated architecture performs no better than simply knowing yesterday’s load-shedding stage—a result the author reports directly rather than burying. The honest interpretation is that infrastructure stage information is the dominant signal in this setting, and no architecture tested, including TENSORnet, exceeds it. What the architecture does provide is a validated, market-observable proxy for stage information in settings where official grid data is delayed, incomplete, or unavailable—a nontrivial value proposition for regulators in jurisdictions where grid operators publish megawatt-shortfall forecasts rather than discrete ordinal stages.

The ablation analysis nonetheless shows that the physics-inspired components are internally load-bearing. Removing the gate collapses the AUC from 0.9994 to 0.568—essentially chance level—demonstrating that unmodulated cross-entropy carries almost no discriminatory power. Substituting Shannon entropy with a rolling volatility measure, while keeping the gate, reduces F1 by several percentage points, indicating that informational diversity specifically, not just any modulated risk proxy, contributes meaningful signal. Against purely market-based benchmarks, the architecture dominates decisively: GARCH volatility models achieve an AUC of only 0.622, and network-connectedness measures reach just 0.516. Under matched small-sample conditions—with only 19 positive training examples at the narrower Stage 4+ definition—TENSORnet’s F1 of 99.7 percent also exceeds a gradient-boosted XGBoost benchmark’s 91.8 percent, a result consistent with the documented inability of tree-based learners to extrapolate beyond their training data.

The study’s most robust independent finding is the confirmation of what the author terms the densification paradox. Network theory predicts that under infrastructure stress, cross-asset correlation should rise while entropy falls—a structural inversion that conventional risk metrics, which treat rising correlation as a sign of diversification and stability, would misread as improvement. The empirical evidence is striking: the correlation between network density and entropy is −0.468 at Stage 4+ versus only −0.184 at Stage 0, with p < 0.001. Crucially, a permutation test against a null distribution of randomly matched day subsets places the true Stage 4+ correlation more than five standard deviations beyond the mechanical-artefact mean, confirming that the inversion is a genuine stage-specific phenomenon rather than an artefact of the two measures sharing overlapping rolling-window information. For stress tests calibrated on normal-conditions correlation data, the implication is sobering: infrastructure-induced fragility would be systematically underestimated precisely when it matters most.

Not everything survives honest testing. The study proposed a cross-scale divergence signal as an advance-warning mechanism—a way of detecting metabolic arrest before it manifests in prices. Evaluated on the held-out test period, the signal fired 20 times; 19 of those episodes were followed by no fragile onset within a 30-day window, a 95 percent false alarm rate. The single genuine test-period onset, the May 2022 Stage 4 escalation, was preceded by only a 4-day lead—far shorter than earlier versions of the study had claimed. The author reports this failure plainly, noting that the signal’s strong serial autocorrelation produces clustered, non-independent threshold crossings rather than independent trials, and that a revised specification requiring sustained activation remains future work. A theoretical port-Hamiltonian neural ODE extension, which would ground metabolic arrest in the second law of thermodynamics, is likewise presented as an unvalidated conceptual pathway rather than an empirical result.

The limitations are disclosed with unusual candor. The VIX proxy used for gate calibration is only modestly correlated with infrastructure stages, and an alternative calibration using JSE realised volatility produced a physically implausible, sign-inverted gate that collapsed to chance-level performance—suggesting that finding a defensible throughput proxy is itself nontrivial. Formal data requests to Eskom for operational engineering telemetry and to the JSE for settlement throughput data went unanswered, and both sources carry structural coverage restrictions that would prevent them from covering the study’s calibration window regardless. The bond-return series underlying one attempted trading-strategy backtest was found to be distorted by stale quotes during high-stress periods, and that comparison was withdrawn rather than reported. Concurrent shocks over the study period—the COVID-19 pandemic, energy price volatility, interest-rate cycles—were not separately parcelled out, a limitation the author states explicitly.

What emerges is less a new early-warning system than a methodological exemplar: a demonstration of how to test, rather than assume, whether a physics-inspired architecture adds value beyond simpler alternatives, and how to report the answer honestly when it complicates the original hypothesis. The open dataset—the only publicly available pairing of daily financial returns with official infrastructure stage labels at decade-long resolution—invites replication in other jurisdictions where infrastructure stress and multi-asset markets coexist, from the Texas grid during the February 2021 winter storm to Europe during the 2022 energy crisis. As climate change intensifies infrastructure disruptions worldwide, the question TENSORnet poses—whether markets’ informational fabric frays in thermodynamically predictable ways when the lights go out—will only grow in urgency.

Subject of Research: Thermodynamically motivated entropy-based detection of infrastructure-induced systemic risk in cross-asset financial networks during South Africa's load-shedding crisis

Article Title: TENSORnet: a thermodynamically motivated entropy protocol for infrastructure-induced metabolic arrest detection in cross-asset financial networks

Article References: Moroke, N. D. (2026). TENSORnet: a thermodynamically motivated entropy protocol for infrastructure-induced metabolic arrest detection in cross-asset financial networks. Frontiers in Physics, 14, Article 1943577. https://doi.org/10.3389/fphy.2026.1943577

Image Credits: AI Generated

DOI: 10.3389/fphy.2026.1943577

Keywords: TENSORnet, Landauer's principle, Shannon entropy, systemic risk, load-shedding, Johannesburg Stock Exchange, financial networks, metabolic arrest, densification paradox, physics-informed machine learning, Eskom, macroprudential policy

The study, authored by Riad Mollik Babu, introduces the concept of a grid-unreliability regime, a statistical description of the interruption process captured by five independent quantities: the fraction of hours per year the grid is down, the mean duration of each outage, the variability of those durations, the predictability of outages in advance, and the timing bias of outages relative to solar availability. Rather than optimizing a single installation under one assumed condition, the analysis swept this five-dimensional space across 1,536 sampled regime points, sizing a distributed photovoltaic and battery system at each one against three objectives: net present cost, expected unserved critical energy, and life-cycle carbon intensity. The work is anchored to Dhaka, Bangladesh, where recurring load shedding exemplifies chronic grid unreliability.

To make results comparable across sites, the designs were expressed as two dimensionless ratios: a storage ratio, defined as battery energy per unit of daily critical load, and a PV ratio, defined as daily solar energy per unit of daily critical load. Outages were generated by an alternating-renewal process with durations drawn from a Gamma distribution, which allows the mean and the spread of outage durations to be set independently, a flexibility the conventional two-state Markov model lacks because it fixes the duration variability at one. A rule-based rolling-horizon controller operated each simulated year at hourly resolution, with predictability entering as the fraction of outage events revealed to the controller before they occur.

The variance decomposition, computed with second-order Sobol indices on the mean of five replicated optimizer runs, produced a striking hierarchy. The outage time fraction dominates both design ratios, with a total-order Sobol index of roughly 0.58 for each. Mean outage duration and duration variability follow. For the storage ratio, the response is almost purely additive: how much, how long, and how variably the grid is out together explain nearly all of the variance. Notably, the storage requirement grows fastest at low outage fractions, meaning the biggest design sensitivity sits in mildly unreliable grids rather than the most severe ones, a regime of diminishing structural return as outages intensify.

The most surprising result concerns predictability. Intuition would place advance knowledge of outages on the storage decision, since a controller that knows an outage is coming can pre-charge the battery. The evidence moves it almost entirely off that decision. Once residual stochastic variance from the optimizer and the synthetic outage years was removed, predictability’s total-order index on the storage ratio fell to about 0.02. What actually lowers the required PV capacity is not foresight but permission: the simple ability to charge the battery from the grid between outages. A blanket pre-charging policy holding no outage information whatsoever captured 87 to 109 percent of the PV reduction achieved by perfect foresight, and the paired difference between the two was not statistically resolved in any tested regime.

Predictability does matter, but through a different channel. At a fixed charging rule, moving from uninformed blanket pre-charging to full foresight cut annual imported grid energy by 67 percent in the severe regime and lowered life-cycle carbon intensity by 45 percent, from 145.6 to 79.6 grams of CO2-equivalent per kilowatt-hour. The advantage shrinks where outages fall after dark, because evening interruptions carry little solar overlap and the battery must be filled largely from the grid whether or not the controller anticipates them. Foresight, in other words, governs the efficiency of grid charging, not the size of the array.

The study also quantified the price of reliability. Median net present cost rose 51 percent from a 95 percent to a 99 percent critical-load availability target, and a further 32 percent to reach 99.9 percent. Every one of the 1,536 sampled regimes proved feasible even at 99.9 percent energy-based availability, demonstrating that chronic load shedding does not impose the survival limit of a rare multi-day outage. The framework was benchmarked against conventional sizing rules with sobering results: sizing the same Dhaka consumer as though the grid were absent inflated cost by 103 to 185 percent, while a days-of-autonomy rule overshot by 156 to 270 percent, because both ignore the energy the grid delivers between interruptions.

Robustness was tested extensively. The design rules transferred across a second solar year, an alternative double-peaked load profile, and two other cities, Karachi and Lagos, with the mean PV ratio moving by under 4 percent and the sensitivity structure statistically indistinguishable between sites. Correlated regime combinations, imperfect forecasts, and narrowed parameter ranges were also examined. When forecast skill degraded from perfect to useless, systems whose pre-charging depended on announced schedules saw PV requirements balloon by 68 to 196 percent, because the controller stopped importing and reverted to solar-only operation. The robust policy is to permit grid charging unconditionally and use any published schedule only to reduce imported energy.

The practical implications reach beyond engineering. Metering or tariff rules that block grid charging of home batteries quietly raise the capacity customers must install, while a reliable load-shedding schedule does not shrink that capacity but does cut the grid energy and emissions needed to hold a given reliability level. Both benefits remain invisible to design methods that treat predictability as a storage-side parameter. By reducing grid unreliability to five measurable quantities and showing how each shapes the cost-optimal design, the study offers a transferable playbook for the billions of people whose grids are present but perpetually imperfect.

Subject of Research: Optimal sizing of distributed PV and battery systems under chronic grid unreliability regimes

Article Title: Resilience value of distributed PV and battery systems under grid-unreliability regimes

Article References: Babu, R. M. (2026). Resilience value of distributed PV and battery systems under grid-unreliability regimes. Results in Engineering, 32, Article 113342. https://doi.org/10.1016/j.rineng.2026.113342

Image Credits: AI Generated

DOI: 10.1016/j.rineng.2026.113342

Keywords: distributed PV, battery storage, grid unreliability, load shedding, resilience, Sobol sensitivity analysis, multi-objective optimization, Dhaka, energy access, predictability, techno-economic sizing, life-cycle carbon